MeloTTS / app.py
cocktailpeanut's picture
update
d798cbb
raw
history blame
2.08 kB
import gradio as gr
import os, torch, io
#os.system('python -m unidic download')
from melo.api import TTS
speed = 1.0
import tempfile
if torch.cuda.is_available():
device = "cuda"
elif torch.backends.mps.is_available():
device = "mps"
else:
device = "cpu"
#device = 'cuda' if torch.cuda.is_available() else 'cpu'
languages = ["EN", "ES", "FR", "ZH", "JP", "KR"]
en = ["EN-Default", "EN-US", "EN-BR", "EN_INDIA", "EN-AU"]
#model = TTS(language='EN', device=device)
def synthesize(language, speaker, text, speed=1.0, progress=gr.Progress()):
model = TTS(language=language, device=device)
speaker_ids = model.hps.data.spk2id
bio = io.BytesIO()
model.tts_to_file(text, speaker_ids[speaker], bio, speed=speed, pbar=progress.tqdm, format='wav')
return bio.getvalue()
def lang(language):
if language == "EN":
return gr.update(choices=en, value="EN-Default")
else:
return gr.update(choices=[language], value=language)
with gr.Blocks() as demo:
gr.Markdown('# MeloTTS\n\nAn unofficial demo of [MeloTTS](https://github.com/myshell-ai/MeloTTS) from MyShell AI. MeloTTS is a permissively licensed (MIT) SOTA multi-speaker TTS model.\n\nI am not affiliated with MyShell AI in any way.\n\nThis demo currently only supports English, but the model itself supports other languages.')
with gr.Group():
language = gr.Dropdown(languages, interactive=True, value='EN', label='Language')
speaker = gr.Dropdown(en, interactive=True, value='EN-Default', label='Speaker')
speed = gr.Slider(label='Speed', minimum=0.1, maximum=10.0, value=1.0, interactive=True, step=0.1)
text = gr.Textbox(label="Text to speak", value='The field of text to speech has seen rapid development recently')
language.change(fn=lang, inputs=[language], outputs=[speaker])
btn = gr.Button('Synthesize', variant='primary')
aud = gr.Audio(interactive=False)
btn.click(synthesize, inputs=[language, speaker, text, speed], outputs=[aud])
demo.queue(api_open=False, default_concurrency_limit=10).launch(show_api=False)