fix: proper Seed-VC generator handling and audio output

app/services/seed_vc_service.py

@@ -107,8 +107,8 @@ class SeedVCService:
 
         try:
             # Perform voice conversion using the wrapper
-            # Returns numpy array of audio samples
-            audio_array = self.wrapper.convert_voice(
+            # Returns generator that yields tuples, final result is in StopIteration.value
+            result_gen = self.wrapper.convert_voice(
                 source=source_audio_path,
                 target=reference_audio_path,
                 diffusion_steps=diffusion_steps,
@@ -120,6 +120,27 @@ class SeedVCService:
                 stream_output=False  # We want the full audio array, not streaming
             )
             
+            # Iterate through generator to get final result
+            # The final return value is accessible via StopIteration exception
+            audio_array = None
+            try:
+                while True:
+                    item = next(result_gen)
+                    # item is (mp3_bytes, full_audio) tuple
+                    if item and len(item) >= 2 and item[1] is not None:
+                        # full_audio is (sr, audio_array) tuple
+                        full_audio = item[1]
+                        if isinstance(full_audio, tuple) and len(full_audio) == 2:
+                            audio_array = full_audio[1]
+                        else:
+                            audio_array = full_audio
+            except StopIteration as e:
+                # The return value of the generator is in e.value
+                if e.value is not None:
+                    # e.value is the numpy array (returned from convert_voice)
+                    audio_array = e.value
+                logger.info(f"Voice conversion complete, audio shape: {audio_array.shape if audio_array is not None else None}")
+
             # Determine sample rate based on f0_condition
             sr = 44100 if f0_condition else 22050
 
@@ -129,6 +150,20 @@ class SeedVCService:
                 # If wrapper returns tuple, extract the array
                 audio_array = audio_array[0]
             
+            # Handle generator - get all chunks and concatenate
+            if hasattr(audio_array, '__iter__') and not isinstance(audio_array, (np.ndarray, list, tuple)):
+                # It's a generator, collect all chunks
+                chunks = []
+                for chunk in audio_array:
+                    if isinstance(chunk, np.ndarray):
+                        chunks.append(chunk)
+                    else:
+                        chunks.append(np.array(chunk))
+                if chunks:
+                    audio_array = np.concatenate(chunks)
+                else:
+                    audio_array = np.array([])
+            
             # Ensure numpy array
             audio_array = np.asarray(audio_array)
             
@@ -148,6 +183,10 @@ class SeedVCService:
             # Ensure float32 format for soundfile
             audio_array = audio_array.astype(np.float32)
 
+            # Convert to absolute path
+            output_path = os.path.abspath(output_path)
+            os.makedirs(os.path.dirname(output_path), exist_ok=True)
+
             # Save output
             sf.write(output_path, audio_array, sr)
             logger.info(f"Voice conversion complete: {output_path}")