Update app.py

app.py

@@ -7,6 +7,7 @@ from zonos.conditioning import make_cond_dict
 
 # Global cache to hold the loaded model
 MODEL = None
+device = "cuda"
 
 def load_model():
     """
@@ -50,14 +51,14 @@ def tts(text, speaker_audio):
     # Get speaker embedding
     with torch.no_grad():
         spk_embedding = model.make_speaker_embedding(wav_tensor, sr)
-        spk_embedding = spk_embedding.to(model.device, dtype=torch.bfloat16)
+        spk_embedding = spk_embedding.to(device, dtype=torch.bfloat16)
 
     # Prepare conditioning dictionary
     cond_dict = make_cond_dict(
         text=text,                # The text prompt
         speaker=spk_embedding,    # Speaker embedding from reference audio
         language="en-us",         # Hard-coded language or switch to another if needed
-        device=model.device,
+        device=device,
     )
     conditioning = model.prepare_conditioning(cond_dict)