Update app.py

app.py

@@ -150,7 +150,7 @@ class NeuTTSWrapper:
         # 3. Infer full text
         with torch.no_grad():
             audio = self.tts_model.infer(text, ref_s, reference_text)
-        return audio.cpu().numpy()
+        return audio
 
     def stream_speech_blocking(self, text: str, ref_audio_path: str, reference_text: str, speed: float, audio_format: str) -> Generator[bytes, None, None]:
         """Sentence-by-Sentence Streaming (Blocking)."""
@@ -175,7 +175,7 @@ class NeuTTSWrapper:
                 audio_chunk = self.tts_model.infer(sentence, ref_s, reference_text)
             
             # Convert and yield
-            yield self._convert_to_streamable_format(audio_chunk.cpu().numpy(), audio_format)
+            yield self._convert_to_streamable_format(audio_chunk, audio_format)
             
         logger.info("Streaming synthesis complete.")