app.py

import torch
from transformers import pipeline
import gradio as gr

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

# Supported multilingual TTS models
tts_lang_model = {
    "yor": "facebook/mms-tts-yor",
    "ibo": "facebook/mms-tts-ibo",
    "hau": "facebook/mms-tts-hau"
}

# Cache of TTS pipelines
tts_pipelines = {}

def get_tts_pipeline(lang_code):
    if lang_code not in tts_pipelines:
        print(f"Loading TTS model for language: {lang_code}")
        tts_pipelines[lang_code] = pipeline(
            "text-to-speech",
            model=tts_lang_model[lang_code],
            device=0 if device == "cuda" else -1
        )
    return tts_pipelines[lang_code]

# Inference function
def synthesize(text, language):
    if not text.strip():
        return None
    tts = get_tts_pipeline(language)
    result = tts(text)
    return (16000, result["audio"])

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🗣️ Multilingual TTS for Nigerian Languages (Yoruba, Igbo, Hausa)")

    with gr.Row():
        lang = gr.Radio(
            choices=["yor", "ibo", "hau"],
            label="Select Language",
            value="yor"
        )

    input_text = gr.Textbox(label="Enter text to synthesize", placeholder="Type something in the selected language")
    output_audio = gr.Audio(label="Generated Speech", type="numpy")

    generate_btn = gr.Button("🔊 Generate")

    generate_btn.click(fn=synthesize, inputs=[input_text, lang], outputs=output_audio)

# Launch interface
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)