app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import soundfile as sf

model_name = "facebook/musicgen-small"
model = AutoModelForCausalLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

def text_to_audio(prompt):
    input_ids = tokenizer(prompt, return_tensors="pt").input_ids
    output = model.generate(input_ids)
    audio_data = output[0].cpu().numpy()
    sf.write("generated_audio.wav", audio_data, 22050)
    return "generated_audio.wav"

gr.Interface(fn=text_to_audio, inputs="text", outputs="audio").launch()