app.py

import gradio as gr
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
import scipy.io.wavfile
from pydub import AudioSegment

# Load the pre-trained model and tokenizer
model_name = "facebook/mms-tts-tam"
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

def audio_to_waveform(audio_file):
    audio = AudioSegment.from_file(audio_file)
    waveform = torch.FloatTensor(audio.get_array_of_samples()).view(1, -1)
    return waveform

def change_voice(input_audio, voice_sample, language):
    # Convert audio files to waveforms
    input_waveform = audio_to_waveform(input_audio)
    voice_waveform = audio_to_waveform(voice_sample)
    
    # Generate the new voice waveform
    text = tokenizer.decode(model.generate(input_waveform))
    inputs = tokenizer(text, return_tensors="pt")
    with torch.no_grad():
        output = model(**inputs).waveform

    # Save to output file
    output_path = "output.wav"
    scipy.io.wavfile.write(output_path, rate=model.config.sampling_rate, data=output.numpy())
    return output_path

def toggle(choice):
    if choice == "mic":
        return gr.update(visible=True, value=None), gr.update(visible=False, value=None)
    else:
        return gr.update(visible=False, value=None), gr.update(visible=True, value=None)

with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            input_audio = gr.Audio(label="Input Audio", type="filepath")
            voice_sample = gr.Audio(label="Voice Sample", type="filepath")
            language = gr.Radio(label="Language", choices=["ta"], value="ta")
            btn = gr.Button("Submit")
        with gr.Column():
            output_audio = gr.Audio(label="Output Audio")

    btn.click(change_voice, inputs=[input_audio, voice_sample, language], outputs=output_audio)

demo.launch()