app.py

import os
import gradio as gr
from TTS.api import TTS
import torch
import spaces # Required for Hugging Face GPU Zero

# 1. Setup Device (Hugging Face ZeroGPU handles this via @spaces.GPU)
device = "cuda" if torch.cuda.is_available() else "cpu"

# 2. Load the Model (XTTS v2 is the gold standard)
# We do this outside the function so it only loads once
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)

@spaces.GPU # This gives you free GPU access on Hugging Face
def clone_voice(text, audio_file):
    if audio_file is None:
        return None, "Please upload a reference audio file."
    
    output_path = "output.wav"
    
    # The actual cloning process
    tts.tts_to_file(
        text=text,
        speaker_wav=audio_file,
        language="en",
        file_path=output_path
    )
    
    return output_path

# 3. Create the Interface
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🎙️ Pro Voice Cloner")
    gr.Markdown("Upload a 6-10 second clip of a voice and type what you want it to say.")
    
    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(label="Text to Speak", placeholder="Hello, I am your cloned voice...", lines=3)
            input_audio = gr.Audio(label="Reference Audio (Clone from this)", type="filepath")
            submit_btn = gr.Button("Clone Voice", variant="primary")
        
        with gr.Column():
            output_audio = gr.Audio(label="Generated Audio")

    submit_btn.click(
        fn=clone_voice,
        inputs=[input_text, input_audio],
        outputs=[output_audio]
    )

demo.launch()