Update app.py

app.py

@@ -154,7 +154,25 @@ if os.path.dirname(os.path.abspath("Amphion")) not in sys.path:
 os.makedirs("wav", exist_ok=True)
 os.makedirs("ckpts/Vevo", exist_ok=True)
 
-from models.vc.vevo.vevo_utils import VevoInferencePipeline, save_audio, load_wav
+# IMPORTANT: Do NOT import save_audio from vevo_utils because it uses torchaudio.save which crashes
+from models.vc.vevo.vevo_utils import VevoInferencePipeline, load_wav
+
+# Define a custom save_audio function using soundfile directly to avoid TorchCodec errors
+def my_save_audio(waveform, output_path, sample_rate=24000):
+    try:
+        # Move to CPU and detach
+        if isinstance(waveform, torch.Tensor):
+            waveform = waveform.detach().cpu()
+            # Handle shapes [1, T] -> [T]
+            if waveform.dim() == 2 and waveform.shape[0] == 1:
+                waveform = waveform.squeeze(0)
+            waveform = waveform.numpy()
+        
+        sf.write(output_path, waveform, sample_rate)
+        print(f"Audio saved successfully to {output_path}")
+    except Exception as e:
+        print(f"Failed to save audio with soundfile: {e}")
+        raise e
 
 # Download and setup config files
 def setup_configs():
@@ -623,8 +641,8 @@ def vevo_style(content_wav, style_wav):
         
         print(f"Generated audio shape: {gen_audio.shape}, max: {torch.max(gen_audio)}, min: {torch.min(gen_audio)}")
         
-        # Save generated audio
-        save_audio(gen_audio, output_path=output_path)
+        # Save generated audio using custom function
+        my_save_audio(gen_audio, output_path=output_path)
         
         return output_path
     except Exception as e:
@@ -717,8 +735,8 @@ def vevo_timbre(content_wav, reference_wav):
         
         print(f"Generated audio shape: {gen_audio.shape}, max: {torch.max(gen_audio)}, min: {torch.min(gen_audio)}")
         
-        # Save generated audio
-        save_audio(gen_audio, output_path=output_path)
+        # Save generated audio using custom function
+        my_save_audio(gen_audio, output_path=output_path)
         
         return output_path
     except Exception as e:
@@ -839,8 +857,8 @@ def vevo_voice(content_wav, style_reference_wav, timbre_reference_wav):
         
         print(f"Generated audio shape: {gen_audio.shape}, max: {torch.max(gen_audio)}, min: {torch.min(gen_audio)}")
         
-        # Save generated audio
-        save_audio(gen_audio, output_path=output_path)
+        # Save generated audio using custom function
+        my_save_audio(gen_audio, output_path=output_path)
         
         return output_path
     except Exception as e:
@@ -942,8 +960,8 @@ def vevo_tts(text, ref_wav, timbre_ref_wav=None, style_ref_text=None, src_langua
         
         print(f"Generated audio shape: {gen_audio.shape}, max: {torch.max(gen_audio)}, min: {torch.min(gen_audio)}")
         
-        # Save generated audio
-        save_audio(gen_audio, output_path=output_path)
+        # Save generated audio using custom function
+        my_save_audio(gen_audio, output_path=output_path)
         
         return output_path
     except Exception as e: