app.py

import gradio as gr
from transformers import pipeline
import scipy.io.wavfile
import numpy as np
import os
import time

# Load the MMS-TTS model for Tibetan
model_id = "ganga4364/mms-tts-bod-female"
synthesiser = pipeline("text-to-speech", model_id)

# Create output directory
os.makedirs("outputs", exist_ok=True)

def generate_audio(input_text):
    if not input_text or not input_text.strip():
        return None, None
    
    # Perform TTS inference
    speech = synthesiser(input_text)
    audio_data = speech["audio"][0]
    sample_rate = speech["sampling_rate"]
    
    # Normalize audio to 16-bit PCM
    audio_data = (audio_data / np.max(np.abs(audio_data)) * 32767).astype(np.int16)
    
    # Create unique filename with timestamp
    timestamp = int(time.time())
    file_path = f"outputs/tibetan_tts_{timestamp}.wav"
    
    # Save the audio file
    scipy.io.wavfile.write(file_path, rate=sample_rate, data=audio_data)
    
    # Return path for both audio player and download
    return file_path, file_path

# Create Gradio interface with download button
with gr.Blocks(title="Tibetan TTS") as demo:
    gr.Markdown("# 🎙️ Tibetan Text-to-Speech")
    gr.Markdown("Enter Tibetan text to generate speech audio with downloadable output.")
    
    with gr.Row():
        text_input = gr.Textbox(
            label="Tibetan Text",
            placeholder="བཀྲ་ཤིས་བདེ་ལེགས།",
            lines=3
        )
    
    generate_btn = gr.Button("🔊 Generate Audio", variant="primary")
    
    with gr.Row():
        audio_output = gr.Audio(label="🎧 Listen to Audio", type="filepath")
    
    with gr.Row():
        download_output = gr.File(label="📥 Download Audio File")
    
    generate_btn.click(
        fn=generate_audio,
        inputs=text_input,
        outputs=[audio_output, download_output]
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)