app.py

import gradio as gr
import os, torch, io
import sys
#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"

languages = ["EN", "ES", "FR", "ZH", "JP", "KR"]
en = ["EN-Default", "EN-US", "EN-BR", "EN_INDIA", "EN-AU"]

LANG = sys.argv[1]

def synthesize(speaker, text, speed=1.0, progress=gr.Progress()):
    model = TTS(language=LANG, 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()

with gr.Blocks() as demo:
    with gr.Group():
        if LANG == "EN":
            speaker = gr.Dropdown(en, interactive=True, value='EN-Default', label='Speaker')
        else:
            speaker = gr.Dropdown([LANG], interactive=True, value=LANG, 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')
    btn = gr.Button('Synthesize', variant='primary')
    aud = gr.Audio(interactive=False)
    btn.click(synthesize, inputs=[speaker, text, speed], outputs=[aud])
demo.queue(api_open=False, default_concurrency_limit=10).launch(show_api=False)