app.py

import gradio as gr
import torch
from transformers import VitsModel, VitsTokenizer

# Load the MMS-TTS model and tokenizer from Hugging Face
MODEL_NAME = "facebook/mms-tts-tam"
tokenizer = VitsTokenizer.from_pretrained(MODEL_NAME)
model = VitsModel.from_pretrained(MODEL_NAME)

# Set up device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

def synthesize_speech(text):
    try:
        if not text.strip():
            raise ValueError("Text input cannot be empty")
        
        # Tokenize input text
        inputs = tokenizer(text, return_tensors="pt").to(device)
        
        # Generate speech
        with torch.no_grad():
            speech = model(**inputs).waveform.cpu().squeeze().numpy()
        
        # Return sample rate and waveform
        sample_rate = model.config.sampling_rate
        return (sample_rate, speech)
    
    except Exception as e:
        return f"Error: {str(e)}", None

# Create Gradio interface
interface = gr.Interface(
    fn=synthesize_speech,
    inputs=gr.Textbox(
        label="Input Text",
        placeholder="Enter text to synthesize...",
        lines=3
    ),
    outputs=gr.Audio(
        label="Generated Speech",
        type="numpy"
    ),
    title="MMS-TTS English Text-to-Speech",
    description="Convert text to speech using Facebook's MMS-TTS-ENG model",
    examples=[
        ["Hello! This is a text-to-speech demonstration."],
        ["The quick brown fox jumps over the lazy dog."],
        ["Natural language processing is fascinating!"]
    ]
)

# Launch the application
if __name__ == "__main__":
    interface.launch(server_name="0.0.0.0" if torch.cuda.is_available() else None)