Checkpoint and traing details

[CleyLyChen]

Hi! Great work! Hello! I am learning about your work, but I cannot achieve the performance mentioned in your paper. I noticed that the gt_mask in NetWrapper is always a tensor with all elements set to 1. Could you provide more training details and your model checkpoint?

[WikiChao]

Hi,

gt_mask is used for training the separation branch. Usually, we have two ways to set gt_masks: (1) binary mask, where the gt_masks represent whether certain Time-Frequency bins are dominated by one of the sounds or the mixture, referred to as gt_masks = (spec_clean > 0.5 * spec).float(). The values should not be all one, unless you choose spec_clean and spec to be the same or the added sound is much smaller than the original one. Besides, you can try different loss terms for penalizing, such as binary loss or L1/L2 loss; (2) ratio mask is also commonly used, with the form of gt_masks = spec_clean / spec. The same loss terms are applied.

Please check if you have mixed another sound at the beginning, if you don't need the separation branch, consider removing the loss term.

[CleyLyChen]

Hi, I removed the separation branch and loss term, and train the model on Epic-Kitchens-100, but its CiOU@0.2 is only 15.2, CiOU@0.3 is 7 and CiOU@0.4 is 1.8, i dont know why, can you share more details?

[WikiChao]

I uploaded a data filtering script, to help remove some silent video clips in the training set. As the training set is constructed using the action recognition benchmark, there are many silent videos, which yield a negative impact on the training process. We tried a simple way to remove some silent videos (please refer to the code), but there could be better ways to explore.

Also, early stopping might be a useful trick, as the model uses a pre-trained vision network as initialization. You can check results from early epochs/steps, to decide whether your model's training is on the right way.

[CleyLyChen]

Hi, I still have some questions during training.

1. I removed the slient video clips according to your scripts, training set is only 9411 video clips left and testing set is only 2898 video clips left.
2. I trained your model for 40 epochs on filtered training set, the best epoch's ciou@0.2 is 18.26, ciou@0.3 is 9.21, ciou@0.4 is 3.86(threshold is median of heatmap), here is my training log, the batch size is 16 and i removed the err.mean() during training.

   Epoch: [1][0/409], Time: 9.45, Data: 5.52, lr_sound: 0.001, lr_frame: 0.0001, loss: 4.1758
   Epoch: [1][20/409], Time: 3.75, Data: 0.32, lr_sound: 0.001, lr_frame: 0.0001, loss: 3.4631
   Epoch: [1][40/409], Time: 2.91, Data: 0.19, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.5754
   Epoch: [1][60/409], Time: 2.83, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.0339
   Epoch: [1][80/409], Time: 3.04, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.8445
   Epoch: [1][100/409], Time: 3.16, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.6860
   Epoch: [1][0/588], Time: 11.46, Data: 7.43, lr_sound: 0.001, lr_frame: 0.0001, loss: 4.1810
   Epoch: [1][20/588], Time: 3.89, Data: 0.41, lr_sound: 0.001, lr_frame: 0.0001, loss: 3.4935
   Epoch: [1][40/588], Time: 3.69, Data: 0.24, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.5676
   Epoch: [1][60/588], Time: 3.63, Data: 0.18, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.2132
   Epoch: [1][80/588], Time: 3.60, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.7622
   Epoch: [1][100/588], Time: 3.57, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.6586
   Epoch: [1][120/588], Time: 3.56, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.4217
   Epoch: [1][140/588], Time: 3.55, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.4287
   Epoch: [1][160/588], Time: 3.55, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.3390
   Epoch: [1][180/588], Time: 3.55, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.1895
   Epoch: [1][200/588], Time: 3.54, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.1199
   Epoch: [1][220/588], Time: 3.53, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8974
   Epoch: [1][240/588], Time: 3.53, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.0010
   Epoch: [1][260/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.0835
   Epoch: [1][280/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.1133
   Epoch: [1][300/588], Time: 3.53, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9576
   Epoch: [1][320/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8214
   Epoch: [1][340/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9479
   Epoch: [1][360/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8073
   Epoch: [1][380/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7141
   Epoch: [1][400/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8135
   Epoch: [1][420/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7217
   Epoch: [1][440/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6688
   Epoch: [1][460/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6411
   Epoch: [1][480/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6243
   Epoch: [1][500/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6810
   Epoch: [1][520/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5658
   Epoch: [1][540/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7880
   Epoch: [1][560/588], Time: 3.52, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5949
   Epoch: [1][580/588], Time: 3.51, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7155
   Epoch: [1][0/588], Time: 11.39, Data: 5.37, lr_sound: 0.001, lr_frame: 0.0001, loss: 4.1810
   Epoch: [1][20/588], Time: 4.09, Data: 0.31, lr_sound: 0.001, lr_frame: 0.0001, loss: 3.4920
   Epoch: [1][40/588], Time: 3.97, Data: 0.19, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.5649
   Epoch: [1][60/588], Time: 3.88, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 2.1807
   Epoch: [1][80/588], Time: 3.83, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.7942
   Epoch: [1][100/588], Time: 3.79, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.5984
   Epoch: [1][120/588], Time: 3.81, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.5884
   Epoch: [1][140/588], Time: 3.79, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.4185
   Epoch: [1][160/588], Time: 3.78, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.2574
   Epoch: [1][180/588], Time: 3.77, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.3440
   Epoch: [1][200/588], Time: 3.77, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.2705
   Epoch: [1][220/588], Time: 3.76, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.0430
   Epoch: [1][240/588], Time: 3.75, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.1693
   Epoch: [1][260/588], Time: 3.75, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9548
   Epoch: [1][280/588], Time: 3.76, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8652
   Epoch: [1][300/588], Time: 3.75, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9799
   Epoch: [1][320/588], Time: 3.75, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9064
   Epoch: [1][340/588], Time: 3.76, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9438
   Epoch: [1][360/588], Time: 3.76, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8222
   Epoch: [1][380/588], Time: 3.76, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7039
   Epoch: [1][400/588], Time: 3.77, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6734
   Epoch: [1][420/588], Time: 3.77, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7620
   Epoch: [1][440/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6265
   Epoch: [1][460/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7539
   Epoch: [1][480/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6029
   Epoch: [1][500/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6748
   Epoch: [1][520/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5724
   Epoch: [1][540/588], Time: 3.78, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5763
   Epoch: [1][560/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5508
   Epoch: [1][580/588], Time: 3.77, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6539
   Epoch: [2][0/588], Time: 10.45, Data: 6.30, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5689
   Epoch: [2][20/588], Time: 4.00, Data: 0.36, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5990
   Epoch: [2][40/588], Time: 3.88, Data: 0.21, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5450
   Epoch: [2][60/588], Time: 3.83, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5504
   Epoch: [2][80/588], Time: 3.79, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5666
   Epoch: [2][100/588], Time: 3.77, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6176
   Epoch: [2][120/588], Time: 3.76, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6018
   Epoch: [2][140/588], Time: 3.75, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5471
   Epoch: [2][160/588], Time: 3.74, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5686
   Epoch: [2][180/588], Time: 3.74, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6198
   Epoch: [2][200/588], Time: 3.74, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.0275
   Epoch: [2][220/588], Time: 3.74, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7355
   Epoch: [2][240/588], Time: 3.74, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7617
   Epoch: [2][260/588], Time: 3.73, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9432
   Epoch: [2][280/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6631
   Epoch: [2][300/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6630
   Epoch: [2][320/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7054
   Epoch: [2][340/588], Time: 3.72, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5094
   Epoch: [2][360/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6511
   Epoch: [2][380/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7991
   Epoch: [2][400/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5536
   Epoch: [2][420/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4725
   Epoch: [2][440/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5614
   Epoch: [2][460/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5916
   Epoch: [2][480/588], Time: 3.73, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4753
   Epoch: [2][500/588], Time: 3.73, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5111
   Epoch: [2][520/588], Time: 3.74, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5404
   Epoch: [2][540/588], Time: 3.74, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5138
   Epoch: [2][560/588], Time: 3.74, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5061
   Epoch: [2][580/588], Time: 3.74, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4305
   Epoch: [3][0/588], Time: 12.32, Data: 7.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5470
   Epoch: [3][20/588], Time: 4.15, Data: 0.40, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5444
   Epoch: [3][40/588], Time: 3.92, Data: 0.24, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4655
   Epoch: [3][60/588], Time: 3.83, Data: 0.18, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6225
   Epoch: [3][80/588], Time: 3.36, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4974
   Epoch: [3][100/588], Time: 3.08, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7689
   Epoch: [3][120/588], Time: 3.17, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5072
   Epoch: [3][140/588], Time: 3.25, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4875
   Epoch: [3][160/588], Time: 3.30, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5146
   Epoch: [3][180/588], Time: 3.34, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5556
   Epoch: [3][200/588], Time: 3.36, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4513
   Epoch: [3][220/588], Time: 3.39, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4138
   Epoch: [3][240/588], Time: 3.42, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6241
   Epoch: [3][260/588], Time: 3.43, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5607
   Epoch: [3][280/588], Time: 3.46, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6141
   Epoch: [3][300/588], Time: 3.47, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5482
   Epoch: [3][320/588], Time: 3.48, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4780
   Epoch: [3][340/588], Time: 3.50, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4840
   Epoch: [3][360/588], Time: 3.51, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4358
   Epoch: [3][380/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5462
   Epoch: [3][400/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5667
   Epoch: [3][420/588], Time: 3.53, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6208
   Epoch: [3][440/588], Time: 3.53, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5835
   Epoch: [3][460/588], Time: 3.54, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4604
   Epoch: [3][480/588], Time: 3.55, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4659
   Epoch: [3][500/588], Time: 3.55, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4573
   Epoch: [3][520/588], Time: 3.55, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6221
   Epoch: [3][540/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4710
   Epoch: [3][560/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4532
   Epoch: [3][580/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4506
   Epoch: [4][0/588], Time: 10.55, Data: 6.33, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4392
   Epoch: [4][20/588], Time: 3.96, Data: 0.36, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4195
   Epoch: [4][40/588], Time: 3.80, Data: 0.22, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4902
   Epoch: [1][0/588], Time: 13.00, Data: 7.71, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4899
   Epoch: [1][20/588], Time: 4.06, Data: 0.43, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5651
   Epoch: [1][40/588], Time: 3.80, Data: 0.25, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5583
   Epoch: [1][60/588], Time: 3.71, Data: 0.19, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4681
   Epoch: [1][80/588], Time: 3.68, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4563
   Epoch: [1][100/588], Time: 3.67, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4984
   Epoch: [1][120/588], Time: 3.65, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5317
   Epoch: [1][140/588], Time: 3.64, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5651
   Epoch: [1][160/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4764
   Epoch: [1][180/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6429
   Epoch: [1][200/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5813
   Epoch: [1][220/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4645
   Epoch: [1][240/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5310
   Epoch: [1][260/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4372
   Epoch: [1][280/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4687
   Epoch: [1][300/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4925
   Epoch: [1][320/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4976
   Epoch: [1][340/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4829
   Epoch: [1][360/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5029
   Epoch: [1][380/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4449
   Epoch: [1][400/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5103
   Epoch: [1][420/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5083
   Epoch: [1][440/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5028
   Epoch: [1][460/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4942
   Epoch: [1][480/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5026
   Epoch: [1][500/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4861
   Epoch: [1][520/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3890
   Epoch: [1][540/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6259
   Epoch: [1][560/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4007
   Epoch: [1][580/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4087
   Epoch: [2][0/588], Time: 10.37, Data: 6.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3901
   Epoch: [2][20/588], Time: 3.90, Data: 0.35, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5464
   Epoch: [2][40/588], Time: 3.75, Data: 0.21, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4392
   Epoch: [2][60/588], Time: 3.70, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4178
   Epoch: [2][80/588], Time: 3.68, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4418
   Epoch: [2][100/588], Time: 3.65, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4693
   Epoch: [2][120/588], Time: 3.63, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4793
   Epoch: [2][140/588], Time: 3.63, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5529
   Epoch: [2][160/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4476
   Epoch: [2][180/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5183
   Epoch: [2][200/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3991
   Epoch: [2][220/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5588
   Epoch: [2][240/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5231
   Epoch: [2][260/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4619
   Epoch: [2][280/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4306
   Epoch: [2][300/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4238
   Epoch: [2][320/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4297
   Epoch: [2][340/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4767
   Epoch: [2][360/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3697
   Epoch: [2][380/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6044
   Epoch: [2][400/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4399
   Epoch: [2][420/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4914
   Epoch: [2][440/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4851
   Epoch: [2][460/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5587
   Epoch: [2][480/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5352
   Epoch: [2][500/588], Time: 3.52, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4812
   Epoch: [2][520/588], Time: 3.48, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5102
   Epoch: [2][540/588], Time: 3.48, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4389
   Epoch: [2][560/588], Time: 3.49, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5080
   Epoch: [2][580/588], Time: 3.49, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5662
   Epoch: [3][0/588], Time: 11.09, Data: 5.81, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9355
   Epoch: [3][20/588], Time: 4.02, Data: 0.34, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5677
   Epoch: [3][40/588], Time: 3.82, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7611
   Epoch: [3][60/588], Time: 3.73, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7395
   Epoch: [3][80/588], Time: 3.70, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7692
   Epoch: [3][100/588], Time: 3.68, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5468
   Epoch: [3][120/588], Time: 3.66, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5703
   Epoch: [3][140/588], Time: 3.66, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7253
   Epoch: [3][160/588], Time: 3.64, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6027
   Epoch: [3][180/588], Time: 3.64, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5449
   Epoch: [3][200/588], Time: 3.63, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6291
   Epoch: [3][220/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4519
   Epoch: [3][240/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6092
   Epoch: [3][260/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5218
   Epoch: [3][280/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4292
   Epoch: [3][300/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5298
   Epoch: [3][320/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4253
   Epoch: [3][340/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5528
   Epoch: [3][360/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5641
   Epoch: [3][380/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5031
   Epoch: [3][400/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4240
   Epoch: [3][420/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6612
   Epoch: [3][440/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5816
   Epoch: [3][460/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5122
   Epoch: [3][480/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4329
   Epoch: [3][500/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5064
   Epoch: [3][520/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5358
   Epoch: [3][540/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5695
   Epoch: [3][560/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4869
   Epoch: [3][580/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3943
   Epoch: [4][0/588], Time: 11.69, Data: 7.50, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5344
   Epoch: [4][20/588], Time: 4.11, Data: 0.47, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4860
   Epoch: [4][40/588], Time: 3.85, Data: 0.27, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4284
   Epoch: [4][60/588], Time: 3.77, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4747
   Epoch: [4][80/588], Time: 3.72, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6278
   Epoch: [4][100/588], Time: 3.69, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4910
   Epoch: [4][120/588], Time: 3.69, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4600
   Epoch: [4][140/588], Time: 3.70, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9107
   Epoch: [4][160/588], Time: 3.68, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6278
   Epoch: [4][180/588], Time: 3.67, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6042
   Epoch: [4][200/588], Time: 3.66, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5486
   Epoch: [4][220/588], Time: 3.66, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4943
   Epoch: [4][240/588], Time: 3.65, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4674
   Epoch: [4][260/588], Time: 3.64, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4376
   Epoch: [4][280/588], Time: 3.64, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5288
   Epoch: [4][300/588], Time: 3.63, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5231
   Epoch: [4][320/588], Time: 3.63, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8455
   Epoch: [4][340/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.9007
   Epoch: [4][360/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 1.3321
   Epoch: [4][380/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7563
   Epoch: [4][400/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7966
   Epoch: [4][420/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5455
   Epoch: [4][440/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7893
   Epoch: [4][460/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8344
   Epoch: [4][480/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6783
   Epoch: [4][500/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6623
   Epoch: [4][520/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5764
   Epoch: [4][540/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4946
   Epoch: [4][560/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7310
   Epoch: [4][580/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5261
   Epoch: [5][0/588], Time: 12.67, Data: 8.52, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5352
   Epoch: [5][20/588], Time: 3.98, Data: 0.47, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5590
   Epoch: [5][40/588], Time: 3.83, Data: 0.27, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4754
   Epoch: [5][60/588], Time: 3.77, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7409
   Epoch: [5][80/588], Time: 3.72, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6370
   Epoch: [5][100/588], Time: 3.68, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7033
   Epoch: [5][120/588], Time: 3.65, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7528
   Epoch: [5][140/588], Time: 3.64, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5645
   Epoch: [5][160/588], Time: 3.63, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5184
   Epoch: [5][180/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4757
   Epoch: [5][200/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4939
   Epoch: [5][220/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5752
   Epoch: [5][240/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6898
   Epoch: [5][260/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7349
   Epoch: [5][280/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5455
   Epoch: [5][300/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3914
   Epoch: [5][320/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4838
   Epoch: [5][340/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5296
   Epoch: [5][360/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4718
   Epoch: [5][380/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4364
   Epoch: [5][400/588], Time: 3.50, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4866
   Epoch: [5][420/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4586
   Epoch: [5][440/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4475
   Epoch: [5][460/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4107
   Epoch: [5][480/588], Time: 3.47, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4221
   Epoch: [5][500/588], Time: 3.47, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4310
   Epoch: [5][520/588], Time: 3.48, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4239
   Epoch: [5][540/588], Time: 3.48, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4278
   Epoch: [5][560/588], Time: 3.48, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3951
   Epoch: [5][580/588], Time: 3.48, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4610
   Epoch: [6][0/588], Time: 11.59, Data: 6.73, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5101
   Epoch: [6][20/588], Time: 3.92, Data: 0.38, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3706
   Epoch: [6][40/588], Time: 3.75, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4121
   Epoch: [6][60/588], Time: 3.68, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4006
   Epoch: [6][80/588], Time: 3.65, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4331
   Epoch: [6][100/588], Time: 3.63, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3715
   Epoch: [6][120/588], Time: 3.67, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4309
   Epoch: [6][140/588], Time: 3.65, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4765
   Epoch: [6][160/588], Time: 3.64, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4571
   Epoch: [6][180/588], Time: 3.63, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4732
   Epoch: [6][200/588], Time: 3.63, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4529
   Epoch: [6][220/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4397
   Epoch: [6][240/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4313
   Epoch: [6][260/588], Time: 3.62, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4684
   Epoch: [6][280/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4346
   Epoch: [6][300/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4327
   Epoch: [6][320/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3990
   Epoch: [6][340/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4235
   Epoch: [6][360/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6646
   Epoch: [6][380/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4443
   Epoch: [6][400/588], Time: 3.65, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4655
   Epoch: [6][420/588], Time: 3.66, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4172
   Epoch: [6][440/588], Time: 3.66, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4796
   Epoch: [6][460/588], Time: 3.67, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4421
   Epoch: [6][480/588], Time: 3.67, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4276
   Epoch: [6][500/588], Time: 3.69, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5559
   Epoch: [6][520/588], Time: 3.69, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4342
   Epoch: [6][540/588], Time: 3.69, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4225
   Epoch: [6][560/588], Time: 3.70, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4076
   Epoch: [6][580/588], Time: 3.70, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4301
   Epoch: [7][0/588], Time: 11.20, Data: 6.86, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4106
   Epoch: [7][20/588], Time: 3.93, Data: 0.39, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3630
   Epoch: [7][40/588], Time: 3.75, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5152
   Epoch: [7][60/588], Time: 3.69, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4418
   Epoch: [7][80/588], Time: 3.66, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4756
   Epoch: [7][100/588], Time: 3.64, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4041
   Epoch: [7][120/588], Time: 3.63, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4156
   Epoch: [7][140/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5060
   Epoch: [7][160/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4319
   Epoch: [7][180/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3917
   Epoch: [7][200/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5006
   Epoch: [7][220/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4999
   Epoch: [7][240/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4131
   Epoch: [7][260/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4138
   Epoch: [7][280/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4626
   Epoch: [7][300/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4632
   Epoch: [7][320/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4477
   Epoch: [7][340/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4730
   Epoch: [7][360/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4873
   Epoch: [7][380/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5394
   Epoch: [7][400/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4647
   Epoch: [7][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6635
   Epoch: [7][440/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6654
   Epoch: [7][460/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4503
   Epoch: [7][480/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3930
   Epoch: [7][500/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4378
   Epoch: [7][520/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4890
   Epoch: [7][540/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4594
   Epoch: [7][560/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3856
   Epoch: [7][580/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3852
   Epoch: [8][0/588], Time: 11.22, Data: 6.87, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3853
   Epoch: [8][20/588], Time: 3.87, Data: 0.39, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4320
   Epoch: [8][40/588], Time: 3.73, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3722
   Epoch: [8][60/588], Time: 3.67, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4052
   Epoch: [8][80/588], Time: 3.65, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4073
   Epoch: [8][100/588], Time: 3.63, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5277
   Epoch: [8][120/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5635
   Epoch: [8][140/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4964
   Epoch: [8][160/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4690
   Epoch: [8][180/588], Time: 3.59, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4031
   Epoch: [8][200/588], Time: 3.59, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5540
   Epoch: [8][220/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4569
   Epoch: [8][240/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4940
   Epoch: [8][260/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4009
   Epoch: [8][280/588], Time: 3.47, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4326
   Epoch: [8][300/588], Time: 3.37, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4601
   Epoch: [8][320/588], Time: 3.39, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8279
   Epoch: [8][340/588], Time: 3.40, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4314
   Epoch: [8][360/588], Time: 3.41, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3750
   Epoch: [8][380/588], Time: 3.42, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3666
   Epoch: [8][400/588], Time: 3.43, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4125
   Epoch: [8][420/588], Time: 3.44, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5684
   Epoch: [8][440/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3692
   Epoch: [8][460/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3926
   Epoch: [8][480/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3957
   Epoch: [8][500/588], Time: 3.47, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4023
   Epoch: [8][520/588], Time: 3.47, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3889
   Epoch: [8][540/588], Time: 3.48, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3953
   Epoch: [8][560/588], Time: 3.49, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3985
   Epoch: [8][580/588], Time: 3.49, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3741
   Epoch: [9][0/588], Time: 10.07, Data: 5.99, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4050
   Epoch: [9][20/588], Time: 3.91, Data: 0.34, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4040
   Epoch: [9][40/588], Time: 3.73, Data: 0.21, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5323
   Epoch: [9][60/588], Time: 3.68, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3872
   Epoch: [9][80/588], Time: 3.65, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4716
   Epoch: [9][100/588], Time: 3.63, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4690
   Epoch: [9][120/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4394
   Epoch: [9][140/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4139
   Epoch: [9][160/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6878
   Epoch: [9][180/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3903
   Epoch: [9][200/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4059
   Epoch: [9][220/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5559
   Epoch: [9][240/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4897
   Epoch: [9][260/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4441
   Epoch: [9][280/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4418
   Epoch: [9][300/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3837
   Epoch: [9][320/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4273
   Epoch: [9][340/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4181
   Epoch: [9][360/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3575
   Epoch: [9][380/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4261
   Epoch: [9][400/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3854
   Epoch: [9][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3686
   Epoch: [9][440/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4349
   Epoch: [9][460/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4734
   Epoch: [9][480/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4291
   Epoch: [9][500/588], Time: 3.59, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4492
   Epoch: [9][520/588], Time: 3.59, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4654
   Epoch: [9][540/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5038
   Epoch: [9][560/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7294
   Epoch: [9][580/588], Time: 3.60, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5328
   Epoch: [10][0/588], Time: 11.46, Data: 6.93, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5151
   Epoch: [10][20/588], Time: 3.90, Data: 0.39, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4711
   Epoch: [10][40/588], Time: 3.73, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4196
   Epoch: [10][60/588], Time: 3.65, Data: 0.18, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4256
   Epoch: [10][80/588], Time: 3.63, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3736
   Epoch: [10][100/588], Time: 3.60, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4482
   Epoch: [10][120/588], Time: 3.59, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4045
   Epoch: [10][140/588], Time: 3.58, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4529
   Epoch: [10][160/588], Time: 3.58, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4945
   Epoch: [10][180/588], Time: 3.57, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5045
   Epoch: [10][200/588], Time: 3.56, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4887
   Epoch: [10][220/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4134
   Epoch: [10][240/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3941
   Epoch: [10][260/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4449
   Epoch: [10][280/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4079
   Epoch: [10][300/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4288
   Epoch: [10][320/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3969
   Epoch: [10][340/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4863
   Epoch: [10][360/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4354
   Epoch: [10][380/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3979
   Epoch: [10][400/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3692
   Epoch: [10][420/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3650
   Epoch: [10][440/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4141
   Epoch: [10][460/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4216
   Epoch: [10][480/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4018
   Epoch: [10][500/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4596
   Epoch: [10][520/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3915
   Epoch: [10][540/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4634
   Epoch: [10][560/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4291
   Epoch: [10][580/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4550
   Epoch: [11][0/588], Time: 11.03, Data: 6.77, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4259
   Epoch: [11][20/588], Time: 3.93, Data: 0.38, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4890
   Epoch: [11][40/588], Time: 3.73, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4168
   Epoch: [11][60/588], Time: 3.66, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4504
   Epoch: [11][80/588], Time: 3.63, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8193
   Epoch: [11][100/588], Time: 3.62, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6823
   Epoch: [11][120/588], Time: 3.60, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7105
   Epoch: [11][140/588], Time: 3.59, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7664
   Epoch: [11][160/588], Time: 3.58, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7094
   Epoch: [11][180/588], Time: 3.43, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5034
   Epoch: [11][200/588], Time: 3.29, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4624
   Epoch: [11][220/588], Time: 3.31, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4042
   Epoch: [11][240/588], Time: 3.34, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4923
   Epoch: [11][260/588], Time: 3.37, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4804
   Epoch: [11][280/588], Time: 3.39, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5170
   Epoch: [11][300/588], Time: 3.40, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5953
   Epoch: [11][320/588], Time: 3.40, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4636
   Epoch: [11][340/588], Time: 3.41, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4575
   Epoch: [11][360/588], Time: 3.42, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4535
   Epoch: [11][380/588], Time: 3.42, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4343
   Epoch: [11][400/588], Time: 3.43, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4597
   Epoch: [11][420/588], Time: 3.43, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5107
   Epoch: [11][440/588], Time: 3.44, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5243
   Epoch: [11][460/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5246
   Epoch: [11][480/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4542
   Epoch: [11][500/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4658
   Epoch: [11][520/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4083
   Epoch: [11][540/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4304
   Epoch: [11][560/588], Time: 3.46, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4246
   Epoch: [11][580/588], Time: 3.46, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4761
   Epoch: [12][0/588], Time: 12.45, Data: 8.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4787
   Epoch: [12][20/588], Time: 3.99, Data: 0.45, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5639
   Epoch: [12][40/588], Time: 3.78, Data: 0.26, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5103
   Epoch: [12][60/588], Time: 3.72, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4421
   Epoch: [12][80/588], Time: 3.67, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5372
   Epoch: [12][100/588], Time: 3.65, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4111
   Epoch: [12][120/588], Time: 3.64, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3956
   Epoch: [12][140/588], Time: 3.63, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4257
   Epoch: [12][160/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5525
   Epoch: [12][180/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4195
   Epoch: [12][200/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4871
   Epoch: [12][220/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4442
   Epoch: [12][240/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4616
   Epoch: [12][260/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3710
   Epoch: [12][280/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3933
   Epoch: [12][300/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5779
   Epoch: [12][320/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4477
   Epoch: [12][340/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4365
   Epoch: [12][360/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4012
   Epoch: [12][380/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4233
   Epoch: [12][400/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4949
   Epoch: [12][420/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4560
   Epoch: [12][440/588], Time: 3.64, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4273
   Epoch: [12][460/588], Time: 3.64, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3979
   Epoch: [12][480/588], Time: 3.65, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5277
   Epoch: [12][500/588], Time: 3.65, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3372
   Epoch: [12][520/588], Time: 3.66, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4502
   Epoch: [12][540/588], Time: 3.66, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5195
   Epoch: [12][560/588], Time: 3.66, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4047
   Epoch: [12][580/588], Time: 3.67, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4309
   Epoch: [13][0/588], Time: 12.70, Data: 8.76, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4487
   Epoch: [13][20/588], Time: 4.00, Data: 0.48, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5673
   Epoch: [13][40/588], Time: 3.78, Data: 0.27, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3961
   Epoch: [13][60/588], Time: 3.71, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4121
   Epoch: [13][80/588], Time: 3.67, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3774
   Epoch: [13][100/588], Time: 3.65, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4860
   Epoch: [13][120/588], Time: 3.63, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3864
   Epoch: [13][140/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3846
   Epoch: [13][160/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3744
   Epoch: [13][180/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5601
   Epoch: [13][200/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3918
   Epoch: [13][220/588], Time: 3.61, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3789
   Epoch: [13][240/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4206
   Epoch: [13][260/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3457
   Epoch: [13][280/588], Time: 3.63, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3727
   Epoch: [13][300/588], Time: 3.64, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3871
   Epoch: [13][320/588], Time: 3.65, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3638
   Epoch: [13][340/588], Time: 3.66, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6451
   Epoch: [13][360/588], Time: 3.66, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4421
   Epoch: [13][380/588], Time: 3.67, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4236
   Epoch: [13][400/588], Time: 3.68, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4055
   Epoch: [13][420/588], Time: 3.68, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4085
   Epoch: [13][440/588], Time: 3.69, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4245
   Epoch: [13][460/588], Time: 3.69, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4899
   Epoch: [13][480/588], Time: 3.70, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6591
   Epoch: [13][500/588], Time: 3.71, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8647
   Epoch: [13][520/588], Time: 3.71, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5140
   Epoch: [13][540/588], Time: 3.71, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4103
   Epoch: [13][560/588], Time: 3.72, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4341
   Epoch: [13][580/588], Time: 3.72, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4560
   Epoch: [14][0/588], Time: 10.88, Data: 6.25, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4225
   Epoch: [14][20/588], Time: 3.86, Data: 0.36, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4947
   Epoch: [14][40/588], Time: 3.16, Data: 0.21, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3784
   Epoch: [14][60/588], Time: 2.84, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3654
   Epoch: [14][80/588], Time: 2.95, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4092
   Epoch: [14][100/588], Time: 3.07, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4060
   Epoch: [14][120/588], Time: 3.15, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4331
   Epoch: [14][140/588], Time: 3.21, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3924
   Epoch: [14][160/588], Time: 3.26, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3616
   Epoch: [14][180/588], Time: 3.28, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3454
   Epoch: [14][200/588], Time: 3.32, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4320
   Epoch: [14][220/588], Time: 3.34, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4455
   Epoch: [14][240/588], Time: 3.36, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4234
   Epoch: [14][260/588], Time: 3.37, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5656
   Epoch: [14][280/588], Time: 3.39, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4093
   Epoch: [14][300/588], Time: 3.40, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4176
   Epoch: [14][320/588], Time: 3.41, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4920
   Epoch: [14][340/588], Time: 3.42, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4483
   Epoch: [14][360/588], Time: 3.43, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4472
   Epoch: [14][380/588], Time: 3.44, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4007
   Epoch: [14][400/588], Time: 3.44, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4546
   Epoch: [14][420/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3748
   Epoch: [14][440/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4257
   Epoch: [14][460/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3861
   Epoch: [14][480/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4747
   Epoch: [14][500/588], Time: 3.46, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4566
   Epoch: [14][520/588], Time: 3.47, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4999
   Epoch: [14][540/588], Time: 3.47, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4948
   Epoch: [14][560/588], Time: 3.47, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4841
   Epoch: [14][580/588], Time: 3.47, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6920
   Epoch: [15][0/588], Time: 10.02, Data: 5.42, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4328
   Epoch: [15][20/588], Time: 3.91, Data: 0.32, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4610
   Epoch: [15][40/588], Time: 3.74, Data: 0.19, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4271
   Epoch: [15][60/588], Time: 3.66, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5275
   Epoch: [15][80/588], Time: 3.63, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5233
   Epoch: [15][100/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5339
   Epoch: [15][120/588], Time: 3.60, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4525
   Epoch: [15][140/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3801
   Epoch: [15][160/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4345
   Epoch: [15][180/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4424
   Epoch: [15][200/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3943
   Epoch: [15][220/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4788
   Epoch: [15][240/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4361
   Epoch: [15][260/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4261
   Epoch: [15][280/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3900
   Epoch: [15][300/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3889
   Epoch: [15][320/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4862
   Epoch: [15][340/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4467
   Epoch: [15][360/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3880
   Epoch: [15][380/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4057
   Epoch: [15][400/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3854
   Epoch: [15][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4939
   Epoch: [15][440/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4149
   Epoch: [15][460/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4335
   Epoch: [15][480/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4548
   Epoch: [15][500/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4816
   Epoch: [15][520/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4577
   Epoch: [15][540/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3465
   Epoch: [15][560/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3807
   Epoch: [15][580/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4127
   Epoch: [16][0/588], Time: 10.62, Data: 6.32, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5604
   Epoch: [16][20/588], Time: 3.92, Data: 0.36, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4247
   Epoch: [16][40/588], Time: 3.74, Data: 0.22, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3837
   Epoch: [16][60/588], Time: 3.71, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3732
   Epoch: [16][80/588], Time: 3.70, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4259
   Epoch: [16][100/588], Time: 3.68, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4514
   Epoch: [16][120/588], Time: 3.66, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4274
   Epoch: [16][140/588], Time: 3.64, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4113
   Epoch: [16][160/588], Time: 3.63, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3748
   Epoch: [16][180/588], Time: 3.62, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4117
   Epoch: [16][200/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3695
   Epoch: [16][220/588], Time: 3.61, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4227
   Epoch: [16][240/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6275
   Epoch: [16][260/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5642
   Epoch: [16][280/588], Time: 3.60, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4409
   Epoch: [16][300/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4046
   Epoch: [16][320/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3976
   Epoch: [16][340/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4038
   Epoch: [16][360/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3935
   Epoch: [16][380/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4028
   Epoch: [16][400/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4029
   Epoch: [16][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4154
   Epoch: [16][440/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4854
   Epoch: [16][460/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5318
   Epoch: [16][480/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4268
   Epoch: [16][500/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.8587
   Epoch: [16][520/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5265
   Epoch: [16][540/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6019
   Epoch: [16][560/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5523
   Epoch: [16][580/588], Time: 3.57, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4825
   Epoch: [17][0/588], Time: 11.03, Data: 6.49, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4139
   Epoch: [17][20/588], Time: 3.89, Data: 0.37, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4928
   Epoch: [17][40/588], Time: 3.70, Data: 0.22, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7734
   Epoch: [17][60/588], Time: 3.64, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5619
   Epoch: [17][80/588], Time: 3.64, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5103
   Epoch: [17][100/588], Time: 3.63, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5089
   Epoch: [17][120/588], Time: 3.62, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5789
   Epoch: [17][140/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5893
   Epoch: [17][160/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4950
   Epoch: [17][180/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6452
   Epoch: [17][200/588], Time: 3.59, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6228
   Epoch: [17][220/588], Time: 3.59, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4331
   Epoch: [17][240/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6075
   Epoch: [17][260/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4189
   Epoch: [17][280/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4356
   Epoch: [17][300/588], Time: 3.58, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4342
   Epoch: [17][320/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5713
   Epoch: [17][340/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5539
   Epoch: [17][360/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4789
   Epoch: [17][380/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4685
   Epoch: [17][400/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6079
   Epoch: [17][420/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5526
   Epoch: [17][440/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5495
   Epoch: [17][460/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4334
   Epoch: [17][480/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4313
   Epoch: [17][500/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4872
   Epoch: [17][520/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4112
   Epoch: [17][540/588], Time: 3.63, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5037
   Epoch: [17][560/588], Time: 3.64, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4349
   Epoch: [17][580/588], Time: 3.64, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4941
   Epoch: [18][0/588], Time: 11.15, Data: 6.80, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4646
   Epoch: [18][20/588], Time: 3.89, Data: 0.38, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3777
   Epoch: [18][40/588], Time: 3.72, Data: 0.23, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4777
   Epoch: [18][60/588], Time: 3.66, Data: 0.17, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5297
   Epoch: [18][80/588], Time: 3.63, Data: 0.15, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3274
   Epoch: [18][100/588], Time: 3.61, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4226
   Epoch: [18][120/588], Time: 3.60, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4254
   Epoch: [18][140/588], Time: 3.60, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4224
   Epoch: [18][160/588], Time: 3.59, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4967
   Epoch: [18][180/588], Time: 3.58, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3812
   Epoch: [18][200/588], Time: 3.58, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6078
   Epoch: [18][220/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4773
   Epoch: [18][240/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4831
   Epoch: [18][260/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7905
   Epoch: [18][280/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7383
   Epoch: [18][300/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3883
   Epoch: [18][320/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4754
   Epoch: [18][340/588], Time: 3.59, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5128
   Epoch: [18][360/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4743
   Epoch: [18][380/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4361
   Epoch: [18][400/588], Time: 3.60, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6699
   Epoch: [18][420/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4068
   Epoch: [18][440/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5126
   Epoch: [18][460/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4245
   Epoch: [18][480/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3979
   Epoch: [18][500/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3890
   Epoch: [18][520/588], Time: 3.61, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4154
   Epoch: [18][540/588], Time: 3.62, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4733
   Epoch: [18][560/588], Time: 3.62, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4525
   Epoch: [18][580/588], Time: 3.62, Data: 0.07, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4519
   Epoch: [19][0/588], Time: 12.19, Data: 8.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3925
   Epoch: [19][20/588], Time: 3.99, Data: 0.45, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4685
   Epoch: [19][40/588], Time: 3.78, Data: 0.26, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4351
   Epoch: [19][60/588], Time: 3.69, Data: 0.20, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3416
   Epoch: [19][80/588], Time: 3.66, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5108
   Epoch: [19][100/588], Time: 3.63, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4229
   Epoch: [19][120/588], Time: 3.62, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3877
   Epoch: [19][140/588], Time: 3.61, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3598
   Epoch: [19][160/588], Time: 3.60, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4814
   Epoch: [19][180/588], Time: 3.59, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4619
   Epoch: [19][200/588], Time: 3.58, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3868
   Epoch: [19][220/588], Time: 3.57, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4044
   Epoch: [19][240/588], Time: 3.57, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3833
   Epoch: [19][260/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3783
   Epoch: [19][280/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3659
   Epoch: [19][300/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4625
   Epoch: [19][320/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.7082
   Epoch: [19][340/588], Time: 3.56, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3851
   Epoch: [19][360/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3693
   Epoch: [19][380/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4211
   Epoch: [19][400/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4087
   Epoch: [19][420/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4090
   Epoch: [19][440/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3891
   Epoch: [19][460/588], Time: 3.54, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3603
   Epoch: [19][480/588], Time: 3.49, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4058
   Epoch: [19][500/588], Time: 3.44, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6536
   Epoch: [19][520/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3765
   Epoch: [19][540/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4769
   Epoch: [19][560/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5043
   Epoch: [19][580/588], Time: 3.45, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6017
   Epoch: [20][0/588], Time: 12.26, Data: 8.02, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4340
   Epoch: [20][20/588], Time: 3.95, Data: 0.44, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4242
   Epoch: [20][40/588], Time: 3.74, Data: 0.26, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4823
   Epoch: [20][60/588], Time: 3.70, Data: 0.19, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3965
   Epoch: [20][80/588], Time: 3.69, Data: 0.16, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4990
   Epoch: [20][100/588], Time: 3.66, Data: 0.14, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4699
   Epoch: [20][120/588], Time: 3.64, Data: 0.13, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4020
   Epoch: [20][140/588], Time: 3.63, Data: 0.12, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4194
   Epoch: [20][160/588], Time: 3.62, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3963
   Epoch: [20][180/588], Time: 3.61, Data: 0.11, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4272
   Epoch: [20][200/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4077
   Epoch: [20][220/588], Time: 3.60, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4344
   Epoch: [20][240/588], Time: 3.59, Data: 0.10, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3984
   Epoch: [20][260/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4088
   Epoch: [20][280/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4674
   Epoch: [20][300/588], Time: 3.58, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4783
   Epoch: [20][320/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4137
   Epoch: [20][340/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3774
   Epoch: [20][360/588], Time: 3.57, Data: 0.09, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4219
   Epoch: [20][380/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3723
   Epoch: [20][400/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4044
   Epoch: [20][420/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5698
   Epoch: [20][440/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4055
   Epoch: [20][460/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4041
   Epoch: [20][480/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4108
   Epoch: [20][500/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.3880
   Epoch: [20][520/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4289
   Epoch: [20][540/588], Time: 3.56, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.4421
   Epoch: [20][560/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.5656
   Epoch: [20][580/588], Time: 3.57, Data: 0.08, lr_sound: 0.001, lr_frame: 0.0001, loss: 0.6244
   Epoch: [21][0/588], Time: 10.36, Data: 5.85, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6479
   Epoch: [21][20/588], Time: 3.86, Data: 0.34, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3788
   Epoch: [21][40/588], Time: 3.72, Data: 0.20, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6105
   Epoch: [21][60/588], Time: 3.65, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3737
   Epoch: [21][80/588], Time: 3.61, Data: 0.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3905
   Epoch: [21][100/588], Time: 3.60, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3670
   Epoch: [21][120/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4420
   Epoch: [21][140/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5857
   Epoch: [21][160/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3873
   Epoch: [21][180/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6601
   Epoch: [21][200/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5735
   Epoch: [21][220/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3563
   Epoch: [21][240/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5797
   Epoch: [21][260/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6502
   Epoch: [21][280/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5655
   Epoch: [21][300/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6812
   Epoch: [21][320/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3992
   Epoch: [21][340/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5721
   Epoch: [21][360/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6239
   Epoch: [21][380/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3453
   Epoch: [21][400/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5431
   Epoch: [21][420/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5803
   Epoch: [21][440/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3768
   Epoch: [21][460/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3649
   Epoch: [21][480/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5527
   Epoch: [21][500/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6654
   Epoch: [21][520/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4141
   Epoch: [21][540/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3871
   Epoch: [21][560/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4290
   Epoch: [21][580/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6501
   Epoch: [22][0/588], Time: 11.04, Data: 6.67, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3636
   Epoch: [22][20/588], Time: 3.88, Data: 0.38, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3656
   Epoch: [22][40/588], Time: 3.70, Data: 0.22, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6042
   Epoch: [22][60/588], Time: 3.64, Data: 0.17, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5815
   Epoch: [22][80/588], Time: 3.62, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5601
   Epoch: [22][100/588], Time: 3.61, Data: 0.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6574
   Epoch: [22][120/588], Time: 3.60, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3657
   Epoch: [22][140/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.7409
   Epoch: [22][160/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6868
   Epoch: [22][180/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5028
   Epoch: [22][200/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5668
   Epoch: [22][220/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6817
   Epoch: [22][240/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5935
   Epoch: [22][260/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6134
   Epoch: [22][280/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6647
   Epoch: [22][300/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3430
   Epoch: [22][320/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3984
   Epoch: [22][340/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4256
   Epoch: [22][360/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5670
   Epoch: [22][380/588], Time: 3.53, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4902
   Epoch: [22][400/588], Time: 3.45, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6103
   Epoch: [22][420/588], Time: 3.43, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3598
   Epoch: [22][440/588], Time: 3.43, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6382
   Epoch: [22][460/588], Time: 3.44, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6018
   Epoch: [22][480/588], Time: 3.44, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4739
   Epoch: [22][500/588], Time: 3.45, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6225
   Epoch: [22][520/588], Time: 3.45, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4441
   Epoch: [22][540/588], Time: 3.46, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3756
   Epoch: [22][560/588], Time: 3.47, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3753
   Epoch: [22][580/588], Time: 3.48, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3608
   Epoch: [23][0/588], Time: 10.69, Data: 6.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3798
   Epoch: [23][20/588], Time: 3.88, Data: 0.36, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3771
   Epoch: [23][40/588], Time: 3.70, Data: 0.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3322
   Epoch: [23][60/588], Time: 3.66, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3418
   Epoch: [23][80/588], Time: 3.63, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3474
   Epoch: [23][100/588], Time: 3.61, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5040
   Epoch: [23][120/588], Time: 3.60, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3326
   Epoch: [23][140/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5006
   Epoch: [23][160/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3584
   Epoch: [23][180/588], Time: 3.60, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3664
   Epoch: [23][200/588], Time: 3.62, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3691
   Epoch: [23][220/588], Time: 3.61, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3462
   Epoch: [23][240/588], Time: 3.61, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3640
   Epoch: [23][260/588], Time: 3.60, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3930
   Epoch: [23][280/588], Time: 3.60, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4322
   Epoch: [23][300/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3862
   Epoch: [23][320/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3674
   Epoch: [23][340/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5060
   Epoch: [23][360/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3910
   Epoch: [23][380/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3941
   Epoch: [23][400/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4842
   Epoch: [23][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3500
   Epoch: [23][440/588], Time: 3.57, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3723
   Epoch: [23][460/588], Time: 3.57, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3568
   Epoch: [23][480/588], Time: 3.57, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4217
   Epoch: [23][500/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4121
   Epoch: [23][520/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3518
   Epoch: [23][540/588], Time: 3.58, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4068
   Epoch: [23][560/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3579
   Epoch: [23][580/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4058
   Epoch: [24][0/588], Time: 10.19, Data: 5.94, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4136
   Epoch: [24][20/588], Time: 3.92, Data: 0.34, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4972
   Epoch: [24][40/588], Time: 3.74, Data: 0.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3708
   Epoch: [24][60/588], Time: 3.70, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3785
   Epoch: [24][80/588], Time: 3.65, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6398
   Epoch: [24][100/588], Time: 3.64, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4020
   Epoch: [24][120/588], Time: 3.62, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3408
   Epoch: [24][140/588], Time: 3.62, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4688
   Epoch: [24][160/588], Time: 3.60, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4631
   Epoch: [24][180/588], Time: 3.60, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3632
   Epoch: [24][200/588], Time: 3.59, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3690
   Epoch: [24][220/588], Time: 3.59, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3802
   Epoch: [24][240/588], Time: 3.59, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4363
   Epoch: [24][260/588], Time: 3.58, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4099
   Epoch: [24][280/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4547
   Epoch: [24][300/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4786
   Epoch: [24][320/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3652
   Epoch: [24][340/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4372
   Epoch: [24][360/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3610
   Epoch: [24][380/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3973
   Epoch: [24][400/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3871
   Epoch: [24][420/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3565
   Epoch: [24][440/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3952
   Epoch: [24][460/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3745
   Epoch: [24][480/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3893
   Epoch: [24][500/588], Time: 3.58, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3428
   Epoch: [24][520/588], Time: 3.58, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3428
   Epoch: [24][540/588], Time: 3.58, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3541
   Epoch: [24][560/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4277
   Epoch: [24][580/588], Time: 3.57, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3831
   Epoch: [25][0/588], Time: 10.09, Data: 5.98, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4383
   Epoch: [25][20/588], Time: 3.87, Data: 0.34, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4794
   Epoch: [25][40/588], Time: 3.70, Data: 0.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4150
   Epoch: [25][60/588], Time: 3.64, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4479
   Epoch: [25][80/588], Time: 3.61, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4144
   Epoch: [25][100/588], Time: 3.60, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4442
   Epoch: [25][120/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4647
   Epoch: [25][140/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3903
   Epoch: [25][160/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3634
   Epoch: [25][180/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4136
   Epoch: [25][200/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3806
   Epoch: [25][220/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3701
   Epoch: [25][240/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4035
   Epoch: [25][260/588], Time: 3.56, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4287
   Epoch: [25][280/588], Time: 3.50, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3839
   Epoch: [25][300/588], Time: 3.41, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3443
   Epoch: [25][320/588], Time: 3.37, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4006
   Epoch: [25][340/588], Time: 3.38, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3616
   Epoch: [25][360/588], Time: 3.39, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4512
   Epoch: [25][380/588], Time: 3.39, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3408
   Epoch: [25][400/588], Time: 3.40, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3678
   Epoch: [25][420/588], Time: 3.40, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3283
   Epoch: [25][440/588], Time: 3.42, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3819
   Epoch: [25][460/588], Time: 3.43, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3440
   Epoch: [25][480/588], Time: 3.44, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4043
   Epoch: [25][500/588], Time: 3.44, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3698
   Epoch: [25][520/588], Time: 3.44, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4062
   Epoch: [25][540/588], Time: 3.45, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3218
   Epoch: [25][560/588], Time: 3.45, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3266
   Epoch: [25][580/588], Time: 3.45, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3361
   Epoch: [26][0/588], Time: 10.03, Data: 5.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3407
   Epoch: [26][20/588], Time: 3.89, Data: 0.31, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3975
   Epoch: [26][40/588], Time: 3.74, Data: 0.19, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4088
   Epoch: [26][60/588], Time: 3.67, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4036
   Epoch: [26][80/588], Time: 3.63, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3672
   Epoch: [26][100/588], Time: 3.60, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3537
   Epoch: [26][120/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4193
   Epoch: [26][140/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4055
   Epoch: [26][160/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5905
   Epoch: [26][180/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3580
   Epoch: [26][200/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3716
   Epoch: [26][220/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3699
   Epoch: [26][240/588], Time: 3.55, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3980
   Epoch: [26][260/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3788
   Epoch: [26][280/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3420
   Epoch: [26][300/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3523
   Epoch: [26][320/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4162
   Epoch: [26][340/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3602
   Epoch: [26][360/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4266
   Epoch: [26][380/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3598
   Epoch: [26][400/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3798
   Epoch: [26][420/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3517
   Epoch: [26][440/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3860
   Epoch: [26][460/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3743
   Epoch: [26][480/588], Time: 3.55, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4346
   Epoch: [26][500/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3484
   Epoch: [26][520/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3438
   Epoch: [26][540/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4046
   Epoch: [26][560/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3803
   Epoch: [26][580/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4078
   Epoch: [27][0/588], Time: 10.58, Data: 5.98, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3676
   Epoch: [27][20/588], Time: 3.92, Data: 0.35, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3323
   Epoch: [27][40/588], Time: 3.72, Data: 0.21, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3875
   Epoch: [27][60/588], Time: 3.67, Data: 0.16, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4119
   Epoch: [27][80/588], Time: 3.63, Data: 0.14, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3776
   Epoch: [27][100/588], Time: 3.61, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4094
   Epoch: [27][120/588], Time: 3.60, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4005
   Epoch: [27][140/588], Time: 3.59, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4217
   Epoch: [27][160/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4083
   Epoch: [27][180/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3481
   Epoch: [27][200/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3332
   Epoch: [27][220/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3504
   Epoch: [27][240/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3227
   Epoch: [27][260/588], Time: 3.56, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3370
   Epoch: [27][280/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3415
   Epoch: [27][300/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4008
   Epoch: [27][320/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3426
   Epoch: [27][340/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4365
   Epoch: [27][360/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3592
   Epoch: [27][380/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5975
   Epoch: [27][400/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3685
   Epoch: [27][420/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3313
   Epoch: [27][440/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4072
   Epoch: [27][460/588], Time: 3.55, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3391
   Epoch: [27][480/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3707
   Epoch: [27][500/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3824
   Epoch: [27][520/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3535
   Epoch: [27][540/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4348
   Epoch: [27][560/588], Time: 3.54, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3988
   Epoch: [27][580/588], Time: 3.53, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3998
   Epoch: [28][0/588], Time: 11.57, Data: 7.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3906
   Epoch: [28][20/588], Time: 3.95, Data: 0.40, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3656
   Epoch: [28][40/588], Time: 3.75, Data: 0.24, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3387
   Epoch: [28][60/588], Time: 3.73, Data: 0.18, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3495
   Epoch: [28][80/588], Time: 3.72, Data: 0.15, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3752
   Epoch: [28][100/588], Time: 3.68, Data: 0.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3451
   Epoch: [28][120/588], Time: 3.66, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3669
   Epoch: [28][140/588], Time: 3.64, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3401
   Epoch: [28][160/588], Time: 3.63, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3942
   Epoch: [28][180/588], Time: 3.62, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3713
   Epoch: [28][200/588], Time: 3.54, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3351
   Epoch: [28][220/588], Time: 3.42, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3635
   Epoch: [28][240/588], Time: 3.34, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4332
   Epoch: [28][260/588], Time: 3.36, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6382
   Epoch: [28][280/588], Time: 3.37, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3738
   Epoch: [28][300/588], Time: 3.38, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3572
   Epoch: [28][320/588], Time: 3.39, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3473
   Epoch: [28][340/588], Time: 3.40, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3714
   Epoch: [28][360/588], Time: 3.40, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3765
   Epoch: [28][380/588], Time: 3.41, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4084
   Epoch: [28][400/588], Time: 3.41, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3933
   Epoch: [28][420/588], Time: 3.42, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3089
   Epoch: [28][440/588], Time: 3.42, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3691
   Epoch: [28][460/588], Time: 3.43, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6418
   Epoch: [28][480/588], Time: 3.43, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3365
   Epoch: [28][500/588], Time: 3.44, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3327
   Epoch: [28][520/588], Time: 3.45, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3637
   Epoch: [28][540/588], Time: 3.46, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3963
   Epoch: [28][560/588], Time: 3.47, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5257
   Epoch: [28][580/588], Time: 3.48, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4325
   Epoch: [29][0/588], Time: 11.05, Data: 6.81, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4593
   Epoch: [29][20/588], Time: 3.87, Data: 0.38, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3301
   Epoch: [29][40/588], Time: 3.70, Data: 0.23, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3600
   Epoch: [29][60/588], Time: 3.63, Data: 0.17, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3581
   Epoch: [29][80/588], Time: 3.61, Data: 0.15, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3555
   Epoch: [29][100/588], Time: 3.58, Data: 0.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3320
   Epoch: [29][120/588], Time: 3.57, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3409
   Epoch: [29][140/588], Time: 3.56, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3512
   Epoch: [29][160/588], Time: 3.56, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4021
   Epoch: [29][180/588], Time: 3.55, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5918
   Epoch: [29][200/588], Time: 3.55, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3602
   Epoch: [29][220/588], Time: 3.55, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3250
   Epoch: [29][240/588], Time: 3.55, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4104
   Epoch: [29][260/588], Time: 3.54, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6283
   Epoch: [29][280/588], Time: 3.54, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3890
   Epoch: [29][300/588], Time: 3.54, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4359
   Epoch: [29][320/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3409
   Epoch: [29][340/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4076
   Epoch: [29][360/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3711
   Epoch: [29][380/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3958
   Epoch: [29][400/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3803
   Epoch: [29][420/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3456
   Epoch: [29][440/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3223
   Epoch: [29][460/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3339
   Epoch: [29][480/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4066
   Epoch: [29][500/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3816
   Epoch: [29][520/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3319
   Epoch: [29][540/588], Time: 3.54, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4031
   Epoch: [29][560/588], Time: 3.53, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3483
   Epoch: [29][580/588], Time: 3.53, Data: 0.07, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3718
   Epoch: [30][0/588], Time: 11.22, Data: 6.87, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3354
   Epoch: [30][20/588], Time: 3.91, Data: 0.39, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3291
   Epoch: [30][40/588], Time: 3.74, Data: 0.23, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3546
   Epoch: [30][60/588], Time: 3.67, Data: 0.17, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3312
   Epoch: [30][80/588], Time: 3.63, Data: 0.15, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3490
   Epoch: [30][100/588], Time: 3.61, Data: 0.13, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3505
   Epoch: [30][120/588], Time: 3.59, Data: 0.12, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3487
   Epoch: [30][140/588], Time: 3.59, Data: 0.11, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4053
   Epoch: [30][160/588], Time: 3.58, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3274
   Epoch: [30][180/588], Time: 3.57, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3829
   Epoch: [30][200/588], Time: 3.56, Data: 0.10, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4055
   Epoch: [30][220/588], Time: 3.58, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3392
   Epoch: [30][240/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.6289
   Epoch: [30][260/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3726
   Epoch: [30][280/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3422
   Epoch: [30][300/588], Time: 3.57, Data: 0.09, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3404
   Epoch: [30][320/588], Time: 3.57, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3824
   Epoch: [30][340/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3670
   Epoch: [30][360/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4178
   Epoch: [30][380/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3650
   Epoch: [30][400/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4549
   Epoch: [30][420/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5066
   Epoch: [30][440/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3211
   Epoch: [30][460/588], Time: 3.58, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3403
   Epoch: [30][480/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3939
   Epoch: [30][500/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4540
   Epoch: [30][520/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3804
   Epoch: [30][540/588], Time: 3.59, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.5974
   Epoch: [30][560/588], Time: 3.60, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.4370
   Epoch: [30][580/588], Time: 3.60, Data: 0.08, lr_sound: 0.0001, lr_frame: 1e-05, loss: 0.3399
   Epoch: [31][0/588], Time: 10.71, Data: 6.42, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4458
   Epoch: [31][20/588], Time: 3.92, Data: 0.37, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3643
   Epoch: [31][40/588], Time: 3.73, Data: 0.22, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3355
   Epoch: [31][60/588], Time: 3.67, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3333
   Epoch: [31][80/588], Time: 3.64, Data: 0.14, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3082
   Epoch: [31][100/588], Time: 3.62, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3776
   Epoch: [31][120/588], Time: 3.42, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3210
   Epoch: [31][140/588], Time: 3.24, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3190
   Epoch: [31][160/588], Time: 3.24, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3640
   Epoch: [31][180/588], Time: 3.27, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3858
   Epoch: [31][200/588], Time: 3.30, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3236
   Epoch: [31][220/588], Time: 3.33, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3156
   Epoch: [31][240/588], Time: 3.37, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3671
   Epoch: [31][260/588], Time: 3.39, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3622
   Epoch: [31][280/588], Time: 3.41, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4114
   Epoch: [31][300/588], Time: 3.42, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3549
   Epoch: [31][320/588], Time: 3.44, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4356
   Epoch: [31][340/588], Time: 3.45, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3631
   Epoch: [31][360/588], Time: 3.46, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4178
   Epoch: [31][380/588], Time: 3.46, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3956
   Epoch: [31][400/588], Time: 3.47, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3890
   Epoch: [31][420/588], Time: 3.47, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3523
   Epoch: [31][440/588], Time: 3.47, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3463
   Epoch: [31][460/588], Time: 3.48, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4847
   Epoch: [31][480/588], Time: 3.48, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4539
   Epoch: [31][500/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3850
   Epoch: [31][520/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3459
   Epoch: [31][540/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3478
   Epoch: [31][560/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3074
   Epoch: [31][580/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3404
   Epoch: [32][0/588], Time: 10.70, Data: 6.14, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3334
   Epoch: [32][20/588], Time: 3.92, Data: 0.36, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.5270
   Epoch: [32][40/588], Time: 3.76, Data: 0.21, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3258
   Epoch: [32][60/588], Time: 3.70, Data: 0.16, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3965
   Epoch: [32][80/588], Time: 3.65, Data: 0.14, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3870
   Epoch: [32][100/588], Time: 3.62, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3409
   Epoch: [32][120/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3277
   Epoch: [32][140/588], Time: 3.61, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3998
   Epoch: [32][160/588], Time: 3.60, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3716
   Epoch: [32][180/588], Time: 3.60, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3506
   Epoch: [32][200/588], Time: 3.59, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3291
   Epoch: [32][220/588], Time: 3.59, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3135
   Epoch: [32][240/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3560
   Epoch: [32][260/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3896
   Epoch: [32][280/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3423
   Epoch: [32][300/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3481
   Epoch: [32][320/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3496
   Epoch: [32][340/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3705
   Epoch: [32][360/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3185
   Epoch: [32][380/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3714
   Epoch: [32][400/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3340
   Epoch: [32][420/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3655
   Epoch: [32][440/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3621
   Epoch: [32][460/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.2996
   Epoch: [32][480/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3565
   Epoch: [32][500/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3735
   Epoch: [32][520/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3286
   Epoch: [32][540/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3190
   Epoch: [32][560/588], Time: 3.57, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3942
   Epoch: [32][580/588], Time: 3.57, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3202
   Epoch: [33][0/588], Time: 12.27, Data: 8.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3580
   Epoch: [33][20/588], Time: 4.03, Data: 0.45, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3605
   Epoch: [33][40/588], Time: 3.80, Data: 0.26, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3770
   Epoch: [33][60/588], Time: 3.72, Data: 0.20, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3572
   Epoch: [33][80/588], Time: 3.67, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3470
   Epoch: [33][100/588], Time: 3.64, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6715
   Epoch: [33][120/588], Time: 3.63, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3759
   Epoch: [33][140/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4220
   Epoch: [33][160/588], Time: 3.62, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3710
   Epoch: [33][180/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4108
   Epoch: [33][200/588], Time: 3.60, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3815
   Epoch: [33][220/588], Time: 3.59, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3612
   Epoch: [33][240/588], Time: 3.59, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3679
   Epoch: [33][260/588], Time: 3.59, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3792
   Epoch: [33][280/588], Time: 3.59, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3153
   Epoch: [33][300/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.5187
   Epoch: [33][320/588], Time: 3.59, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4615
   Epoch: [33][340/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3520
   Epoch: [33][360/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3609
   Epoch: [33][380/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3296
   Epoch: [33][400/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3831
   Epoch: [33][420/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3143
   Epoch: [33][440/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3553
   Epoch: [33][460/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4489
   Epoch: [33][480/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3263
   Epoch: [33][500/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3154
   Epoch: [33][520/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3727
   Epoch: [33][540/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3832
   Epoch: [33][560/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3905
   Epoch: [33][580/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3305
   Epoch: [34][0/588], Time: 11.20, Data: 6.84, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3398
   Epoch: [34][20/588], Time: 3.36, Data: 0.39, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3621
   Epoch: [34][40/588], Time: 2.66, Data: 0.23, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3437
   Epoch: [34][60/588], Time: 2.62, Data: 0.18, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3828
   Epoch: [34][80/588], Time: 2.89, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3262
   Epoch: [34][100/588], Time: 3.05, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3721
   Epoch: [34][120/588], Time: 3.15, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3466
   Epoch: [34][140/588], Time: 3.23, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3633
   Epoch: [34][160/588], Time: 3.28, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3416
   Epoch: [34][180/588], Time: 3.32, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3251
   Epoch: [34][200/588], Time: 3.36, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3617
   Epoch: [34][220/588], Time: 3.39, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4142
   Epoch: [34][240/588], Time: 3.41, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3937
   Epoch: [34][260/588], Time: 3.42, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3582
   Epoch: [34][280/588], Time: 3.44, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3384
   Epoch: [34][300/588], Time: 3.45, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4356
   Epoch: [34][320/588], Time: 3.46, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3542
   Epoch: [34][340/588], Time: 3.47, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3339
   Epoch: [34][360/588], Time: 3.48, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3346
   Epoch: [34][380/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3600
   Epoch: [34][400/588], Time: 3.50, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3118
   Epoch: [34][420/588], Time: 3.50, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3470
   Epoch: [34][440/588], Time: 3.51, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3397
   Epoch: [34][460/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3387
   Epoch: [34][480/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3524
   Epoch: [34][500/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4512
   Epoch: [34][520/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3266
   Epoch: [34][540/588], Time: 3.54, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3696
   Epoch: [34][560/588], Time: 3.54, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3651
   Epoch: [34][580/588], Time: 3.54, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3262
   Epoch: [35][0/588], Time: 12.53, Data: 8.45, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3351
   Epoch: [35][20/588], Time: 3.96, Data: 0.46, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3743
   Epoch: [35][40/588], Time: 3.77, Data: 0.27, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4320
   Epoch: [35][60/588], Time: 3.69, Data: 0.20, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4278
   Epoch: [35][80/588], Time: 3.65, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3385
   Epoch: [35][100/588], Time: 3.64, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3342
   Epoch: [35][120/588], Time: 3.63, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3556
   Epoch: [35][140/588], Time: 3.64, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3002
   Epoch: [35][160/588], Time: 3.64, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3477
   Epoch: [35][180/588], Time: 3.65, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3940
   Epoch: [35][200/588], Time: 3.67, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4446
   Epoch: [35][220/588], Time: 3.66, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3434
   Epoch: [35][240/588], Time: 3.66, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3520
   Epoch: [35][260/588], Time: 3.65, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4343
   Epoch: [35][280/588], Time: 3.64, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3722
   Epoch: [35][300/588], Time: 3.64, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4176
   Epoch: [35][320/588], Time: 3.63, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3487
   Epoch: [35][340/588], Time: 3.63, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3253
   Epoch: [35][360/588], Time: 3.64, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4084
   Epoch: [35][380/588], Time: 3.64, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3325
   Epoch: [35][400/588], Time: 3.65, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3186
   Epoch: [35][420/588], Time: 3.65, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4150
   Epoch: [35][440/588], Time: 3.66, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3139
   Epoch: [35][460/588], Time: 3.66, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3583
   Epoch: [35][480/588], Time: 3.66, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3238
   Epoch: [35][500/588], Time: 3.66, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3295
   Epoch: [35][520/588], Time: 3.66, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3415
   Epoch: [35][540/588], Time: 3.67, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3263
   Epoch: [35][560/588], Time: 3.67, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3509
   Epoch: [35][580/588], Time: 3.67, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3366
   Epoch: [36][0/588], Time: 10.77, Data: 6.42, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3352
   Epoch: [36][20/588], Time: 3.94, Data: 0.37, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3314
   Epoch: [36][40/588], Time: 3.77, Data: 0.22, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3402
   Epoch: [36][60/588], Time: 3.69, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3341
   Epoch: [36][80/588], Time: 3.66, Data: 0.14, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4877
   Epoch: [36][100/588], Time: 3.63, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3673
   Epoch: [36][120/588], Time: 3.62, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3727
   Epoch: [36][140/588], Time: 3.61, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3624
   Epoch: [36][160/588], Time: 3.60, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3266
   Epoch: [36][180/588], Time: 3.60, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3816
   Epoch: [36][200/588], Time: 3.60, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3768
   Epoch: [36][220/588], Time: 3.60, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3788
   Epoch: [36][240/588], Time: 3.59, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3190
   Epoch: [36][260/588], Time: 3.59, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4657
   Epoch: [36][280/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3776
   Epoch: [36][300/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3735
   Epoch: [36][320/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.5781
   Epoch: [36][340/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3493
   Epoch: [36][360/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3581
   Epoch: [36][380/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3131
   Epoch: [36][400/588], Time: 3.59, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3763
   Epoch: [36][420/588], Time: 3.59, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4228
   Epoch: [36][440/588], Time: 3.59, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3265
   Epoch: [36][460/588], Time: 3.60, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3011
   Epoch: [36][480/588], Time: 3.61, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3431
   Epoch: [36][500/588], Time: 3.61, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3653
   Epoch: [36][520/588], Time: 3.62, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3498
   Epoch: [36][540/588], Time: 3.58, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3986
   Epoch: [36][560/588], Time: 3.53, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3949
   Epoch: [36][580/588], Time: 3.52, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3926
   Epoch: [37][0/588], Time: 11.21, Data: 7.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3325
   Epoch: [37][20/588], Time: 3.91, Data: 0.40, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3141
   Epoch: [37][40/588], Time: 3.72, Data: 0.24, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3676
   Epoch: [37][60/588], Time: 3.67, Data: 0.18, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3666
   Epoch: [37][80/588], Time: 3.64, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3070
   Epoch: [37][100/588], Time: 3.62, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4426
   Epoch: [37][120/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3601
   Epoch: [37][140/588], Time: 3.59, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3347
   Epoch: [37][160/588], Time: 3.58, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3554
   Epoch: [37][180/588], Time: 3.58, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3161
   Epoch: [37][200/588], Time: 3.58, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3679
   Epoch: [37][220/588], Time: 3.58, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6818
   Epoch: [37][240/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4143
   Epoch: [37][260/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3414
   Epoch: [37][280/588], Time: 3.56, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3222
   Epoch: [37][300/588], Time: 3.56, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3428
   Epoch: [37][320/588], Time: 3.56, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3892
   Epoch: [37][340/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4390
   Epoch: [37][360/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3207
   Epoch: [37][380/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3385
   Epoch: [37][400/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3980
   Epoch: [37][420/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3076
   Epoch: [37][440/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3518
   Epoch: [37][460/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4140
   Epoch: [37][480/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3549
   Epoch: [37][500/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3890
   Epoch: [37][520/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4008
   Epoch: [37][540/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3388
   Epoch: [37][560/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3563
   Epoch: [37][580/588], Time: 3.57, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3432
   Epoch: [38][0/588], Time: 10.96, Data: 6.66, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3733
   Epoch: [38][20/588], Time: 3.91, Data: 0.38, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3046
   Epoch: [38][40/588], Time: 3.74, Data: 0.23, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3502
   Epoch: [38][60/588], Time: 3.68, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4164
   Epoch: [38][80/588], Time: 3.65, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3567
   Epoch: [38][100/588], Time: 3.63, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3608
   Epoch: [38][120/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3295
   Epoch: [38][140/588], Time: 3.61, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3155
   Epoch: [38][160/588], Time: 3.60, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3477
   Epoch: [38][180/588], Time: 3.59, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3236
   Epoch: [38][200/588], Time: 3.58, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3473
   Epoch: [38][220/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4402
   Epoch: [38][240/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3466
   Epoch: [38][260/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3302
   Epoch: [38][280/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3508
   Epoch: [38][300/588], Time: 3.56, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.5815
   Epoch: [38][320/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3931
   Epoch: [38][340/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4879
   Epoch: [38][360/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3511
   Epoch: [38][380/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4940
   Epoch: [38][400/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3069
   Epoch: [38][420/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4348
   Epoch: [38][440/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3259
   Epoch: [38][460/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6314
   Epoch: [38][480/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3181
   Epoch: [38][500/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4068
   Epoch: [38][520/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3383
   Epoch: [38][540/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3203
   Epoch: [38][560/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3694
   Epoch: [38][580/588], Time: 3.55, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3409
   Epoch: [39][0/588], Time: 11.27, Data: 6.90, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3762
   Epoch: [39][20/588], Time: 3.94, Data: 0.39, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3435
   Epoch: [39][40/588], Time: 3.73, Data: 0.23, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3324
   Epoch: [39][60/588], Time: 3.66, Data: 0.18, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3914
   Epoch: [39][80/588], Time: 3.63, Data: 0.15, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6171
   Epoch: [39][100/588], Time: 3.62, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4659
   Epoch: [39][120/588], Time: 3.61, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3257
   Epoch: [39][140/588], Time: 3.60, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3264
   Epoch: [39][160/588], Time: 3.59, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3570
   Epoch: [39][180/588], Time: 3.59, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3244
   Epoch: [39][200/588], Time: 3.58, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3323
   Epoch: [39][220/588], Time: 3.58, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3436
   Epoch: [39][240/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4165
   Epoch: [39][260/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3451
   Epoch: [39][280/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3121
   Epoch: [39][300/588], Time: 3.57, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3341
   Epoch: [39][320/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3775
   Epoch: [39][340/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3177
   Epoch: [39][360/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3599
   Epoch: [39][380/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3516
   Epoch: [39][400/588], Time: 3.56, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3775
   Epoch: [39][420/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3648
   Epoch: [39][440/588], Time: 3.55, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3435
   Epoch: [39][460/588], Time: 3.49, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3251
   Epoch: [39][480/588], Time: 3.43, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3373
   Epoch: [39][500/588], Time: 3.44, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6936
   Epoch: [39][520/588], Time: 3.45, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3529
   Epoch: [39][540/588], Time: 3.46, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4356
   Epoch: [39][560/588], Time: 3.47, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6890
   Epoch: [39][580/588], Time: 3.47, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3572
   Epoch: [40][0/588], Time: 10.93, Data: 6.57, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.6221
   Epoch: [40][20/588], Time: 3.84, Data: 0.37, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4456
   Epoch: [40][40/588], Time: 3.70, Data: 0.22, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3676
   Epoch: [40][60/588], Time: 3.63, Data: 0.17, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3674
   Epoch: [40][80/588], Time: 3.58, Data: 0.14, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3922
   Epoch: [40][100/588], Time: 3.57, Data: 0.13, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3712
   Epoch: [40][120/588], Time: 3.56, Data: 0.12, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3604
   Epoch: [40][140/588], Time: 3.55, Data: 0.11, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3782
   Epoch: [40][160/588], Time: 3.55, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3133
   Epoch: [40][180/588], Time: 3.54, Data: 0.10, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3467
   Epoch: [40][200/588], Time: 3.53, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3651
   Epoch: [40][220/588], Time: 3.54, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4619
   Epoch: [40][240/588], Time: 3.53, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3811
   Epoch: [40][260/588], Time: 3.53, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.7480
   Epoch: [40][280/588], Time: 3.53, Data: 0.09, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.7117
   Epoch: [40][300/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3294
   Epoch: [40][320/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3547
   Epoch: [40][340/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4467
   Epoch: [40][360/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3892
   Epoch: [40][380/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3190
   Epoch: [40][400/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3575
   Epoch: [40][420/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3312
   Epoch: [40][440/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3201
   Epoch: [40][460/588], Time: 3.53, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4096
   Epoch: [40][480/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3710
   Epoch: [40][500/588], Time: 3.52, Data: 0.08, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.4106
   Epoch: [40][520/588], Time: 3.52, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3141
   Epoch: [40][540/588], Time: 3.52, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3097
   Epoch: [40][560/588], Time: 3.52, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3800
   Epoch: [40][580/588], Time: 3.52, Data: 0.07, lr_sound: 1e-05, lr_frame: 1.0000000000000002e-06, loss: 0.3195

3. here are some visualization cases on testing set(threshold is 0.5): i noticed the result can not focus on sounding object accurately, do you know what's the matter?
4. BTW, can you share your training log and checkpoint to help me debug? Thanks!