CosyVoice-300M / app.py
wenmeng.zwm
update code
f7defef
raw
history blame
3.95 kB
import logging
logging.basicConfig(level=logging.INFO)
import spaces
import os
import sys
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append('{}/third_party/AcademiCodec'.format(ROOT_DIR))
sys.path.append('{}/third_party/Matcha-TTS'.format(ROOT_DIR))
from modelscope import snapshot_download
snapshot_download('speech_tts/speech_kantts_ttsfrd',revision="master", allow_file_pattern='ttsfrd-0.3.9-cp310-cp310-linux_x86_64.whl', local_dir='pretrained_models/speech_kantts_ttsfrd')
snapshot_download('speech_tts/speech_kantts_ttsfrd',revision="master", allow_file_pattern='ttsfrd_dependency-0.1-py3-none-any.whl', local_dir='pretrained_models/speech_kantts_ttsfrd')
os.system('cd pretrained_models/speech_kantts_ttsfrd/ && pip install ttsfrd_dependency-0.1-py3-none-any.whl')
os.system('cd pretrained_models/speech_kantts_ttsfrd/ && pip install ttsfrd-0.3.9-cp310-cp310-linux_x86_64.whl')
os.system('sed -i [email protected]@typing_extensions@g /usr/local/lib/python3.10/site-packages/inflect/__init__.py')
import gradio as gr
from css.advanced import advanced
from css.custom import custom
from css.preset import preset
# audio_mode_choices = [('预置语音生成', 'preset'), ('定制语音生成(复刻录制声音)', 'custom'),
# ('高级语音生成(自然语言控制)', 'advanced')]
# def on_audio_mode_change(_audio_mode_radio):
# yield {
# preset_layout: gr.update(visible=_audio_mode_radio == 'preset'),
# custom_layout: gr.update(visible=_audio_mode_radio == 'custom'),
# advanced_layout: gr.update(visible=_audio_mode_radio == 'advanced')
# }
# custom_css = """
# .full-height {
# height: 100%;
# }
# """
# default_layout = 'preset'
# with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
# audio_mode_radio = gr.Radio(choices=audio_mode_choices,
# value=default_layout,
# label="选择语音生成模式")
# with gr.Row():
# with gr.Column(visible=default_layout == 'preset') as preset_layout:
# preset()
# with gr.Column(visible=default_layout == 'custom') as custom_layout:
# custom()
# with gr.Column(
# visible=default_layout == 'advanced') as advanced_layout:
# advanced()
# audio_mode_radio.change(
# fn=on_audio_mode_change,
# inputs=[audio_mode_radio],
# outputs=[preset_layout, custom_layout, advanced_layout])
audio_mode_choices = [('Preset voice generation', 'preset'), ('Custom voice generation (replicate recorded voice)', 'custom'),
('Advanced voice generation (natural language control)', 'advanced')]
def on_audio_mode_change(_audio_mode_radio):
yield {
preset_layout: gr.update(visible=_audio_mode_radio == 'preset'),
custom_layout: gr.update(visible=_audio_mode_radio == 'custom'),
advanced_layout: gr.update(visible=_audio_mode_radio == 'advanced')
}
custom_css = """
.full-height {
height: 100%;
}
"""
default_layout = 'preset'
with gr.Blocks(theme=gr.themes.Soft(), css=custom_css) as demo:
audio_mode_radio = gr.Radio(choices=audio_mode_choices,
value=default_layout,
label="Select voice generation mode")
with gr.Row():
with gr.Column(visible=default_layout == 'preset') as preset_layout:
preset()
with gr.Column(visible=default_layout == 'custom') as custom_layout:
custom()
with gr.Column(
visible=default_layout == 'advanced') as advanced_layout:
advanced()
audio_mode_radio.change(
fn=on_audio_mode_change,
inputs=[audio_mode_radio],
outputs=[preset_layout, custom_layout, advanced_layout])
#demo.queue().launch(server_port=7860, share=True)
demo.queue().launch(server_port=7860)