update

examples/conv_tasnet/step_2_train_model.py

@@ -195,7 +195,7 @@ def main():
         step_idx = int(step_idx)
         if step_idx > last_step_idx:
             last_step_idx = step_idx
-    last_epoch = 2
+    last_epoch = 1
 
     if last_step_idx != -1:
         logger.info(f"resume from steps-{last_step_idx}.")
@@ -299,7 +299,8 @@ def main():
             # loss = 1.0 * ae_loss + 0.8 * neg_si_snr_loss + 0.5 * mr_stft_loss + 0.3 * neg_stoi_loss
             # loss = 1.0 * ae_loss + 0.8 * neg_si_snr_loss + 0.7 * mr_stft_loss + 0.5 * neg_stoi_loss
             # loss = 2.0 * mr_stft_loss + 0.8 * ae_loss + 0.7 * neg_si_snr_loss + 0.5 * neg_stoi_loss
-            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
+            # 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
+            loss = 0.2 * ae_loss + 0.2 * neg_si_snr_loss + 1.0 * mr_stft_loss + 0.3 * neg_stoi_loss + 0.5 * pesq_loss
             if torch.any(torch.isnan(loss)) or torch.any(torch.isinf(loss)):
                 logger.info(f"find nan or inf in loss.")
                 continue

toolbox/torch/utils/data/dataset/denoise_jsonl_dataset.py

@@ -68,6 +68,7 @@ class DenoiseJsonlDataset(IterableDataset):
             yield self.convert_sample(sample)
 
     def iterable_source(self):
+        last_sample = None
         with open(self.jsonl_file, "r", encoding="utf-8") as f:
             for row in f:
                 row = json.loads(row)
@@ -99,7 +100,11 @@ class DenoiseJsonlDataset(IterableDataset):
 
                     "snr_db": snr_db,
                 }
+                if last_sample is None:
+                    last_sample = sample
+                    continue
                 yield sample
+        yield last_sample
 
     def convert_sample(self, sample: dict):
         noise_filename = sample["noise_filename"]

toolbox/torchaudio/models/frcrn/__init__.py

@@ -0,0 +1,6 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+
+
+if __name__ == "__main__":
+    pass

toolbox/torchaudio/models/frcrn/modeling_frcrn.py

@@ -0,0 +1,10 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+"""
+https://arxiv.org/abs/2206.07293
+"""
+from modelscope.models.audio.ans.frcrn import FRCRN
+
+
+if __name__ == "__main__":
+    pass