Fix WebSocket streaming - ParlerTTSStreamer yields numpy arrays not tensors

api.py

@@ -230,8 +230,8 @@ def generate_audio_chunks_streaming(text, speaker, pitch, rate, temperature, do_
     # Yield audio chunks as they're generated
     for audio_chunk in streamer:
         if audio_chunk.shape[0] > 0:
-            # Convert to int16
-            audio_int16 = (audio_chunk.cpu().numpy() * 32767).astype(np.int16)
+            # ParlerTTSStreamer yields numpy arrays directly, not tensors
+            audio_int16 = (audio_chunk * 32767).astype(np.int16)
             yield audio_int16
 
     thread.join()