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

model_id = "facebook/mms-tts-bod"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = VitsForTextToSpeech.from_pretrained(model_id)

def tts(text):
    inputs = tokenizer(text, return_tensors="pt")
    speech = model.generate_speech(inputs["input_ids"][0], speaker_id=0)
    sf.write("output.wav", speech.cpu().numpy(), samplerate=16000)
    return "output.wav"

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