blisc
commented
4 years ago Why not use both? The learned mel -> mag mapping would be better than mapping mel -> mag by using the transposed mel filter.
It is also possible that the style wav is too different from the training data that it cannot generate a good style token from it.
Hi, I am trying to do inference using pretrained model on single GPU machine. It gets the style right. However, the generated voice is of very bad quality. Please fing attached generated mel diagram and sample voice. I am not able to figure out what is the issue. Do I need to finetune it? https://github.com/gurpreet395/downloads/blob/master/Output_step0_0_infer.png https://github.com/gurpreet395/downloads/blob/master/sample_step0_0_infer.wav
In the config file, I am using "mel" instead of both Here is link to the model. https://drive.google.com/file/d/1IdWnUIwV9NMe-1JSvcv4Ti4HZ8wPEvQr/view
Following is the log from the inference job.
Inference Mode. Loss part of graph isn't built. 2019-11-05 00:40:05.085872: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2499995000 Hz 2019-11-05 00:40:05.086239: I tensorflow/compiler/xla/service/service.cc:161] XLA service 0x6d63b30 executing computations on platform Host. Devices: 2019-11-05 00:40:05.086271: I tensorflow/compiler/xla/service/service.cc:168] StreamExecutor device (0):,
2019-11-05 00:40:05.186390: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2019-11-05 00:40:05.187310: I tensorflow/compiler/xla/service/service.cc:161] XLA service 0x6d65c00 executing computations on platform CUDA. Devices:
2019-11-05 00:40:05.187336: I tensorflow/compiler/xla/service/service.cc:168] StreamExecutor device (0): Tesla T4, Compute Capability 7.5
2019-11-05 00:40:05.187479: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1433] Found device 0 with properties:
name: Tesla T4 major: 7 minor: 5 memoryClockRate(GHz): 1.59
pciBusID: 0000:00:1e.0
totalMemory: 14.75GiB freeMemory: 14.65GiB
2019-11-05 00:40:05.187504: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1512] Adding visible gpu devices: 0
2019-11-05 00:40:05.510047: I tensorflow/core/common_runtime/gpu/gpu_device.cc:984] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-11-05 00:40:05.510091: I tensorflow/core/common_runtime/gpu/gpu_device.cc:990] 0
2019-11-05 00:40:05.510100: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1003] 0: N
2019-11-05 00:40:05.510211: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 14164 MB memory) -> physical GPU (device: 0, name: Tesla T4, pci bus id: 0000:00:1e.0, compute capability: 7.5)
WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/tensorflow/python/util/decorator_utils.py:145: GraphKeys.VARIABLES (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use WARNING: Can't compute number of objects per step, since train model does not define get_num_objects_per_step method.
2019-11-05 00:40:06.334238: I tensorflow/stream_executor/dso_loader.cc:153] successfully opened CUDA library libcublas.so.10 locally
Processed 1/1 batches
Processed 1/1 batches
Not enough steps for benchmarking
output_file is ignored for tts
results are logged to the logdir
Finished inference
tf.GraphKeys.GLOBAL_VARIABLESinstead. WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py:1266: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version. Instructions for updating: Use standard file APIs to check for files with this prefix.