app.py

import gradio as gr
import torch
from transformers import AutoModelForTextToWaveform, AutoTokenizer

model_id = "kyutai/tts-1.6b-en_fr"
model = AutoModelForTextToWaveform.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(model_id)

def tts_fn(text, language="en"):
    inputs = tokenizer(text, return_tensors="pt")
    with torch.no_grad():
        audio = model.generate(**inputs, language=language)
    return (16000, audio.cpu().numpy())

demo = gr.Interface(
    fn=tts_fn,
    inputs=[
        gr.Textbox(label="Enter text"),
        gr.Dropdown(["en", "fr"], value="en", label="Language"),
    ],
    outputs=gr.Audio(label="Generated Speech"),
    title="Kyutai TTS Free Demo (CPU)",
    description="⚠️ Running on free CPU → slow, but works!"
)

if __name__ == "__main__":
    demo.launch()