Update app.py

app.py

@@ -207,7 +207,7 @@ def voice_conversion(input, reference, steps, guidance, speed):
     # Generate Whisper features
     print("[INFO] | Generating Whisper features for source audio.")
     if converted_waves_16k.size(-1) <= sampling_rate * 30:
-        alt_inputs = whisper_feature_extractor([converted_waves_16k.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate)
+        alt_inputs = whisper_feature_extractor([converted_waves_16k.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate=sampling_rate)
         alt_input_features = whisper_model._mask_input_features(alt_inputs.input_features, attention_mask=alt_inputs.attention_mask).to(device)
         alt_outputs = whisper_model.encoder(alt_input_features.to(torch.float32), head_mask=None, output_attentions=False, output_hidden_states=False, return_dict=True)
         S_alt = alt_outputs.last_hidden_state.to(torch.float32)
@@ -229,7 +229,7 @@ def voice_conversion(input, reference, steps, guidance, speed):
                 chunk = converted_waves_16k[:, traversed_time:traversed_time + chunk_size]
             else:
                 chunk = torch.cat([buffer, converted_waves_16k[:, traversed_time:traversed_time + chunk_size - overlap_size]], dim=-1)
-            alt_inputs = whisper_feature_extractor([chunk.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate)
+            alt_inputs = whisper_feature_extractor([chunk.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate=sampling_rate)
             alt_input_features = whisper_model._mask_input_features(alt_inputs.input_features, attention_mask=alt_inputs.attention_mask).to(device)
             alt_outputs = whisper_model.encoder(alt_input_features.to(torch.float32), head_mask=None, output_attentions=False, output_hidden_states=False, return_dict=True)
             S_chunk = alt_outputs.last_hidden_state.to(torch.float32)
@@ -251,7 +251,7 @@ def voice_conversion(input, reference, steps, guidance, speed):
     # Original Whisper features
     print("[INFO] | Generating Whisper features for reference audio.")
     ori_waves_16k = torchaudio.functional.resample(ref_audio_tensor, sr_current, sampling_rate)
-    ori_inputs = whisper_feature_extractor([ori_waves_16k.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate)
+    ori_inputs = whisper_feature_extractor([ori_waves_16k.squeeze(0).cpu().numpy()], return_tensors="pt", return_attention_mask=True, sampling_rate=sampling_rate)
     ori_input_features = whisper_model._mask_input_features(ori_inputs.input_features, attention_mask=ori_inputs.attention_mask).to(device)
     ori_outputs = whisper_model.encoder(ori_input_features.to(torch.float32), head_mask=None, output_attentions=False, output_hidden_states=False, return_dict=True)
     S_ori = ori_outputs.last_hidden_state.to(torch.float32)