fix: handle Seed-VC audio array shape

app/services/seed_vc_service.py

@@ -123,6 +123,31 @@ class SeedVCService:
             # Determine sample rate based on f0_condition
             sr = 44100 if f0_condition else 22050
 
+            # Fix audio array shape for soundfile.write()
+            # Seed-VC may return array with different shapes that need adjustment
+            if isinstance(audio_array, tuple):
+                # If wrapper returns tuple, extract the array
+                audio_array = audio_array[0]
+            
+            # Ensure numpy array
+            audio_array = np.asarray(audio_array)
+            
+            # Handle shape issues
+            if len(audio_array.shape) == 1:
+                # Mono audio, shape is (samples,) - already correct
+                pass
+            elif len(audio_array.shape) == 2:
+                # Could be (channels, samples) or (samples, channels)
+                if audio_array.shape[0] < audio_array.shape[1]:
+                    # Shape is (channels, samples), take first channel
+                    audio_array = audio_array[0]
+                else:
+                    # Shape is (samples, channels), take first channel
+                    audio_array = audio_array[:, 0]
+            
+            # Ensure float32 format for soundfile
+            audio_array = audio_array.astype(np.float32)
+
             # Save output
             sf.write(output_path, audio_array, sr)
             logger.info(f"Voice conversion complete: {output_path}")