update

examples/conv_tasnet/step_2_train_model.py

@@ -144,7 +144,7 @@ def main():
         max_wave_value=32768.0,
         min_snr_db=config.min_snr_db,
         max_snr_db=config.max_snr_db,
-        skip=225000,
+        # skip=225000,
     )
     valid_dataset = DenoiseJsonlDataset(
         jsonl_file=args.valid_dataset,
@@ -195,7 +195,7 @@ def main():
         step_idx = int(step_idx)
         if step_idx > last_step_idx:
             last_step_idx = step_idx
-    last_epoch = 1
+    last_epoch = 2
 
     if last_step_idx != -1:
         logger.info(f"resume from steps-{last_step_idx}.")

examples/conv_tasnet_gan/step_2_train_model.py

@@ -364,7 +364,7 @@ def main():
             neg_si_snr_loss = neg_si_snr_loss_fn.forward(denoise_audios, clean_audios)
             neg_stoi_loss = neg_stoi_loss_fn.forward(denoise_audios, clean_audios)
             mr_stft_loss = mr_stft_loss_fn.forward(denoise_audios, clean_audios)
-            pesq_loss = pesq_loss_fn.forward(clean_audios, denoise_audios)
+            pesq_loss = pesq_loss_fn.forward(denoise_audios, clean_audios)
 
             metric_g = discriminator.forward(denoise_audios, clean_audios)
             discriminator_g_loss = F.mse_loss(metric_g.flatten(), one_labels)
@@ -451,7 +451,7 @@ def main():
                         neg_si_snr_loss = neg_si_snr_loss_fn.forward(denoise_audios, clean_audios)
                         neg_stoi_loss = neg_stoi_loss_fn.forward(denoise_audios, clean_audios)
                         mr_stft_loss = mr_stft_loss_fn.forward(denoise_audios, clean_audios)
-                        pesq_loss = pesq_loss_fn.forward(clean_audios, denoise_audios)
+                        pesq_loss = pesq_loss_fn.forward(denoise_audios, clean_audios)
 
                         loss = 1.0 * ae_loss + 0.8 * neg_si_snr_loss + 0.7 * mr_stft_loss + 0.5 * neg_stoi_loss + 0.5 * pesq_loss
                         if torch.any(torch.isnan(loss)) or torch.any(torch.isinf(loss)):