xVASynth-TTS / app.py
Pendrokar's picture
generate whitenoise
ea6a933
raw
history blame
797 Bytes
import server as xvaserver
import gradio as gr
import requests
def predict(input):
model_type = 'xVAPitch'
line = 'Test'
pace = 1.0
save_path = 'test.wav'
language = 'en'
base_speaker_emb = []
use_sr = 0
use_cleanup = 0
data = {
'modelType': model_type,
'sequence': line,
'pace': pace,
'outfile': save_path,
'vocoder': 'n/a',
'base_lang': language,
'base_emb': base_speaker_emb,
'useSR': use_sr,
'useCleanup': use_cleanup,
}
requests.post('http://localhost:8008/synthesize', json=data)
return ''
input_textbox = gr.Textbox(
label="Input Text",
lines=1,
autofocus=True
)
gradio_app = gr.Interface(
predict,
input_textbox,
outputs="text",
title="xVASynth",
)
if __name__ == "__main__":
gradio_app.launch()