Update app.py

app.py

@@ -4,6 +4,7 @@ from kokoro import KPipeline
 import soundfile as sf
 import os
 import numpy as np
+import torch 
 
 app = FastAPI()
 
@@ -34,11 +35,15 @@ async def generate_audio(text: str, voice: str = "af_heart", speed: float = 1.0)
 
     # Process only the first segment for demo
     for i, (gs, ps, audio) in enumerate(generator):
+
+        # Convert PyTorch tensor to NumPy array
+        audio_numpy = audio.cpu().numpy()
         # Convert to 16-bit PCM
+        
         # Ensure the audio is in the range [-1, 1]
-        audio = np.clip(audio, -1, 1)
+        audio_numpy = np.clip(audio_numpy, -1, 1)
         # Convert to 16-bit signed integers
-        pcm_data = (audio * 32767).astype(np.int16)
+        pcm_data = (audio_numpy * 32767).astype(np.int16)
         
         # Convert to bytes (automatically uses row-major order)
         raw_audio = pcm_data.tobytes()