app.py

import gradio as gr
from transformers import VitsModel, AutoTokenizer
import torch
import numpy as np

model = VitsModel.from_pretrained("facebook/mms-tts-hat")
tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-hat")

def generate_speech(text):
    inputs = tokenizer(text, return_tensors="pt")
    with torch.no_grad():
        output = model(**inputs).waveform
    return (model.config.sampling_rate, output[0].numpy())

demo = gr.Interface(
    fn=generate_speech,
    inputs=gr.Textbox(label="Haitian Creole Text"),
    outputs=gr.Audio(label="Generated Speech"),
    title="Haitian Creole TTS"
)

demo.launch()