update

examples/conv_tasnet/step_2_train_model.py

@@ -204,7 +204,6 @@ def main():
     ae_loss_fn = nn.L1Loss(reduction="mean").to(device)
     neg_si_snr_loss_fn = NegativeSISNRLoss(reduction="mean").to(device)
     neg_stoi_loss_fn = NegSTOILoss(sample_rate=8000, reduction="mean").to(device)
-    lds_loss_fn = LSDLoss(reduction="mean").to(device)
     mr_stft_loss_fn = MultiResolutionSTFTLoss(
         # fft_size_list=[256, 512, 1024],
         win_size_list=[120, 240, 480],
@@ -220,7 +219,6 @@ def main():
     average_ae_loss = 1000000000
     average_neg_si_snr_loss = 1000000000
     average_neg_stoi_loss = 1000000000
-    average_lds_loss = 1000000000
 
     model_list = list()
     best_idx_epoch = None
@@ -237,7 +235,6 @@ def main():
         total_ae_loss = 0.
         total_neg_si_snr_loss = 0.
         total_neg_stoi_loss = 0.
-        total_lds_loss = 0.
         total_batches = 0.
         progress_bar = tqdm(
             desc="Training; epoch-{}".format(idx_epoch),
@@ -256,10 +253,9 @@ def main():
             ae_loss = ae_loss_fn.forward(denoise_audios, clean_audios)
             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)
-            lds_loss = lds_loss_fn.forward(denoise_audios, clean_audios)
             mr_stft_loss = mr_stft_loss_fn.forward(denoise_audios, clean_audios)
 
-            loss = 0.25 * ae_loss + 0.25 * neg_si_snr_loss + 0.25 * neg_stoi_loss + 0.25 * lds_loss + 0.25 * mr_stft_loss
+            loss = 0.25 * ae_loss + 0.25 * neg_si_snr_loss + 0.25 * neg_stoi_loss + 0.25 * mr_stft_loss
 
             denoise_audios_list_r = list(denoise_audios.detach().cpu().numpy())
             clean_audios_list_r = list(clean_audios.detach().cpu().numpy())
@@ -275,7 +271,6 @@ def main():
             total_ae_loss += ae_loss.item()
             total_neg_si_snr_loss += neg_si_snr_loss.item()
             total_neg_stoi_loss += neg_stoi_loss.item()
-            total_lds_loss += lds_loss.item()
             total_batches += 1
 
             average_pesq_score = round(total_pesq_score / total_batches, 4)
@@ -283,7 +278,6 @@ def main():
             average_ae_loss = round(total_ae_loss / total_batches, 4)
             average_neg_si_snr_loss = round(total_neg_si_snr_loss / total_batches, 4)
             average_neg_stoi_loss = round(total_neg_stoi_loss / total_batches, 4)
-            average_lds_loss = round(total_lds_loss / total_batches, 4)
 
             progress_bar.update(1)
             progress_bar.set_postfix({
@@ -292,7 +286,6 @@ def main():
                 "ae_loss": average_ae_loss,
                 "neg_si_snr_loss": average_neg_si_snr_loss,
                 "neg_stoi_loss": average_neg_stoi_loss,
-                "lds_loss": average_lds_loss,
             })
 
         # evaluation
@@ -304,7 +297,6 @@ def main():
         total_ae_loss = 0.
         total_neg_si_snr_loss = 0.
         total_neg_stoi_loss = 0.
-        total_lds_loss = 0.
         total_batches = 0.
 
         progress_bar = tqdm(
@@ -322,9 +314,9 @@ def main():
                 ae_loss = ae_loss_fn.forward(denoise_audios, clean_audios)
                 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)
-                lds_loss = lds_loss_fn.forward(denoise_audios, clean_audios)
+                mr_stft_loss = mr_stft_loss_fn.forward(denoise_audios, clean_audios)
 
-                loss = 0.25 * ae_loss + 0.25 * neg_si_snr_loss + 0.25 * neg_stoi_loss + 0.25 * lds_loss
+                loss = 0.25 * ae_loss + 0.25 * neg_si_snr_loss + 0.25 * neg_stoi_loss + 0.25 * mr_stft_loss
 
                 denoise_audios_list_r = list(denoise_audios.detach().cpu().numpy())
                 clean_audios_list_r = list(clean_audios.detach().cpu().numpy())
@@ -335,7 +327,6 @@ def main():
                 total_ae_loss += ae_loss.item()
                 total_neg_si_snr_loss += neg_si_snr_loss.item()
                 total_neg_stoi_loss += neg_stoi_loss.item()
-                total_lds_loss += lds_loss.item()
                 total_batches += 1
 
                 average_pesq_score = round(total_pesq_score / total_batches, 4)
@@ -343,7 +334,6 @@ def main():
                 average_ae_loss = round(total_ae_loss / total_batches, 4)
                 average_neg_si_snr_loss = round(total_neg_si_snr_loss / total_batches, 4)
                 average_neg_stoi_loss = round(total_neg_stoi_loss / total_batches, 4)
-                average_lds_loss = round(total_lds_loss / total_batches, 4)
 
                 progress_bar.update(1)
                 progress_bar.set_postfix({
@@ -352,7 +342,6 @@ def main():
                     "ae_loss": average_ae_loss,
                     "neg_si_snr_loss": average_neg_si_snr_loss,
                     "neg_stoi_loss": average_neg_stoi_loss,
-                    "lds_loss": average_lds_loss,
                 })
 
         # scheduler
@@ -392,7 +381,6 @@ def main():
             "ae_loss": average_ae_loss,
             "neg_si_snr_loss": average_neg_si_snr_loss,
             "neg_stoi_loss": average_neg_stoi_loss,
-            "lds_loss": average_lds_loss,
         }
         metrics_filename = epoch_dir / "metrics_epoch.json"
         with open(metrics_filename, "w", encoding="utf-8") as f: