update

examples/cnn_vad_by_webrtcvad/step_4_train_model.py

@@ -79,11 +79,11 @@ class CollateFunction(object):
 
         for sample in batch:
             noisy_wave: torch.Tensor = sample["noisy_wave"]
-            speech_wave: torch.Tensor = sample["speech_wave"]
+            clean_wave: torch.Tensor = sample["clean_wave"]
             vad_segments: List[Tuple[float, float]] = sample["vad_segments"]
 
             noisy_audios.append(noisy_wave)
-            clean_audios.append(speech_wave)
+            clean_audios.append(clean_wave)
             batch_vad_segments.append(vad_segments)
 
         noisy_audios = torch.stack(noisy_audios)

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

@@ -151,6 +151,7 @@ class VadPaddingJsonlDataset(IterableDataset):
             snr_db=snr_db, eps=self.eps,
         )
         noisy_wave = torch.tensor(noisy_wave_np, dtype=torch.float32)
+        clean_wave = torch.tensor(speech_wave_np, dtype=torch.float32)
 
         vad_segments = [
             [
@@ -162,7 +163,7 @@ class VadPaddingJsonlDataset(IterableDataset):
 
         result = {
             "noisy_wave": noisy_wave,
-            "speech_wave": speech_wave,
+            "clean_wave": clean_wave,
             "vad_segments": vad_segments,
         }
         return result