Update app.py

app.py

@@ -1,4 +1,4 @@
-# !pip install TTS gradio numpy librosa torch
+# !pip install TTS gradio numpy librosa torch soundfile
 
 from TTS.api import TTS
 import gradio as gr
@@ -7,13 +7,14 @@ import librosa
 import torch
 import tempfile
 import os
+import soundfile as sf  # Added for better audio handling
 
 # Check device availability
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Initialize TTS model
+# Initialize TTS model with device parameter
 model_name = "tts_models/multilingual/multi-dataset/your_tts"
-tts = TTS(model_name=model_name).to(device)
+tts = TTS(model_name=model_name).to(device)  # This line is the problem
 
 def process_audio(audio_path, max_duration=10):
     """Load and trim audio to specified duration"""
@@ -33,7 +34,7 @@ def generate_speech(audio_file, text):
     
     # Process reference audio
     y, sr = process_audio(audio_file)
-    librosa.output.write_wav(ref_path, y, sr)
+    sf.write(ref_path, y, sr)  # Using soundfile instead of librosa for writing
     
     # Generate speech
     try:
@@ -91,7 +92,7 @@ with gr.Blocks(title="Voice Clone TTS") as demo:
         inputs=[audio_input, text_input],
         outputs=audio_output,
         fn=generate_speech,
-        cache_examples=True
+        cache_examples=False  # Disabled cache to avoid potential issues
     )
     
     btn.click(