app.py

import torch
import gradio as gr
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
import soundfile as sf

device = "cuda" if torch.cuda.is_available() else "cpu"

model = AutoModelForSpeechSeq2Seq.from_pretrained("facebook/mms-tts-chv", torch_dtype=torch.float32).to(device)
processor = AutoProcessor.from_pretrained("facebook/mms-tts-chv")

def tts_fn(text):
    inputs = processor(text=text, return_tensors="pt").to(device)
    with torch.no_grad():
        speech = model.generate(**inputs)
    audio = processor.batch_decode(speech, return_tensors="np")[0]
    sf.write("output.wav", audio, 16000)
    return "output.wav"

demo = gr.Interface(fn=tts_fn, inputs="text", outputs="audio", title="TTS Chuvash - Facebook MMS")

demo.launch()