yxlu-0102 / MP-SENet

Explicit Estimation of Magnitude and Phase Spectra in Parallel for High-Quality Speech Enhancement
MIT License
293 stars 44 forks source link

Sample rate (fs) - No default. Must select either 8000 or 16000. #20

Open risqaliyevds opened 7 months ago

risqaliyevds commented 7 months ago

I am training with 48k but in validation I get this

    Run model on reference(ref) and degraded(deg)
    Sample rate (fs) - No default. Must select either 8000 or 16000.
    Note there is narrow band (nb) mode only when sampling rate is 8000Hz.

How to fix it or it is imposible

risqaliyevds commented 7 months ago

I changed all 16k to 48k but i get this eror

yxlu-0102 commented 7 months ago

Could you let me know when you encountered this issue, and which code corresponds to this error? Did it occur during the calculation of PESQ?

risqaliyevds commented 7 months ago

{ "num_gpus": 8, "batch_size": 16, "learning_rate": 0.0005, "adam_b1": 0.8, "adam_b2": 0.99, "lr_decay": 0.99, "seed": 1234,

"dense_channel": 64,
"compress_factor": 0.3, 
"num_tsconformers": 4,
"beta": 2.0,

"sampling_rate": 48000,
"segment_size": 96000,
"n_fft": 800,
"hop_size": 300, 
"win_size": 800, 

"num_workers": 64,

"dist_config": {
    "dist_backend": "nccl",
    "dist_url": "tcp://localhost:54321",
    "world_size": 1
}

}

I set this setttings. I couldn't find this error where located that is why I'm asking.

yxlu-0102 commented 7 months ago

I understand that the metric discriminator needs to compute PESQ during training, but PESQ only supports speech below 16kHz.

yxlu-0102 commented 7 months ago

You can remove the metric discriminator, or find another metric that can be calculated under a 48 kHz sampling rate, such as ViSQOL.

risqaliyevds commented 7 months ago

aha thanks