app.py

import os
os.environ["NUMBA_DISABLE_CACHE"] = "1"
import gradio as gr
import os
import torch

# Add openvoice path
import sys
sys.path.append("openvoice")

from openvoice.api import ToneColorConverter
from openvoice.inference import voice_conversion

# Set up paths
ckpt_converter = './checkpoints/converter'
device = "cuda" if torch.cuda.is_available() else "cpu"

converter = ToneColorConverter(f"{ckpt_converter}/config.json", device=device)
converter.load_ckpt(f"{ckpt_converter}/converter.ckpt")

def convert_voice(audio_file, text_prompt):
    output_path = "./results/output.wav"
    # You must clone reference audio using clone.sh or similar step in Dockerfile
    voice_conversion(converter, audio_file.name, text_prompt, output_path, device)
    return output_path

iface = gr.Interface(
    fn=convert_voice,
    inputs=[
        gr.Audio(type="filepath", label="Input Voice (WAV)"),
        gr.Textbox(label="Prompt (e.g., 'Speak in a cheerful tone')"),
    ],
    outputs=gr.Audio(label="Converted Voice")
)

iface.launch()