Shadhil's picture
user previlages
5457abc
raw
history blame
985 Bytes
import subprocess
# Run the setup.py install command
try:
subprocess.run(['python', 'setup.py', 'install', '--user'], check=True)
print("Installation successful.")
except subprocess.CalledProcessError as e:
print(f"Installation failed with error: {e}")
import gradio as gr
import torch
from TTS.api import TTS
# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"
# Init TTS
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)
def voice_clone(text: str, speaker_wav: str, language: str):
# Run TTS
print("Speaker wav:", speaker_wav)
tts.tts_to_file(text=text, speaker_wav=speaker_wav, language=language, file_path="output.wav")
return "output.wav"
iface = gr.Interface(fn=voice_clone,
inputs=["text", gr.Audio(type="filepath"), gr.Radio(['ko', 'en'], label="Language")],
outputs=gr.Audio(type="filepath"),
title="IMCapusle Voice Clone")
iface.launch()