app.py

import gradio as gr
import torch
from transformers import AutoProcessor, AutoModel
import scipy.io.wavfile as wavfile
import spaces

# Processor
def load_model():
    processor = AutoProcessor.from_pretrained("suno/bark-small")
    model = AutoModel.from_pretrained("suno/bark-small")
    model.eval()  # Set the model to evaluation mode
    return processor, model

# Load models on startup
print("Loading models...")
processor, model = load_model()
print("Models loaded successfully!")

@spaces.GPU  # Decorate the function to enable GPU usage
def text_to_speech(text):
    try:
        # Check if a GPU is available and set device
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        
        # Move model to GPU
        model.to(device)
        
        inputs = processor(
            text=[text],
            return_tensors="pt",
        ).to(device)  # Move inputs to GPU
        
        # Generate speech values on the GPU
        with torch.no_grad():  # Disable gradient calculation for inference
            speech_values = model.generate(**inputs, do_sample=True)
        
        # Move generated audio data back to CPU for saving
        audio_data = speech_values.cpu().numpy().squeeze()
        sampling_rate = model.generation_config.sample_rate
        
        temp_path = "temp_audio.wav"
        wavfile.write(temp_path, sampling_rate, audio_data)
        
        return temp_path
    except Exception as e:
        return f"Error generating speech: {str(e)}"

# Define Gradio interface
demo = gr.Interface(
    fn=text_to_speech,
    inputs=[
        gr.Textbox(
            label="Enter text", 
            placeholder="दिल्ली मेट्रो में आपका स्वागत है"
        )
    ],
    outputs=gr.Audio(label="Generated Speech"),
    title="Bark TTS Test App",
    description="This app generates speech from text using the Bark TTS model.",
    examples=[
        ["दिल्ली मेट्रो में आपका स्वागत है"],
        ["अगला स्टेशन राजीव चौक है"]
    ],
    theme="default"
)

if __name__ == "__main__":
    demo.launch()