Spaces:
Sleeping
Sleeping
import gradio as gr | |
import numpy as np | |
from ipa import g2p | |
from ipa.ipa import text_to_ipa | |
from models import models_config | |
def _do_tts(model_id, ipa, language_name, speaker_name=None, speaker_wav=None): | |
model = models_config[model_id]["model"] | |
if speaker_wav is not None: | |
return model.tts( | |
ipa, | |
speaker_wav=speaker_wav, | |
language_name=language_name, | |
split_sentences=False, | |
) | |
return model.tts( | |
ipa, | |
speaker_name=speaker_name, | |
language_name=language_name, | |
split_sentences=False, | |
) | |
def text_to_speech( | |
model_id: str, | |
use_default_emb_or_custom: str, | |
speaker_wav, | |
speaker: str, | |
language: str, | |
dialect: str, | |
speed: float, | |
text: str, | |
): | |
if len(text) == 0: | |
raise gr.Error("請勿輸入空字串。") | |
tag = language | |
if language not in g2p: | |
tag = f"{language}_{dialect}" | |
ipa = text_to_ipa(text, tag, g2p) | |
models_config[model_id]["model"].tts_model.length_scale = speed | |
if use_default_emb_or_custom == "預設語者": | |
wav = _do_tts( | |
model_id, | |
ipa, | |
speaker_name=speaker | |
if len(models_config[model_id]["speaker_mapping"]) > 1 | |
else None, | |
language_name=language, | |
) | |
else: | |
wav = _do_tts( | |
model_id, | |
ipa, | |
speaker_wav=speaker_wav, | |
language_name=language, | |
) | |
return ( | |
models_config[model_id]["model"].tts_model.config.audio.sample_rate, | |
np.array(wav), | |
) | |
def when_model_selected(model_id): | |
model_config = models_config[model_id] | |
speaker_drop_down_choices = [ | |
(k, v) for k, v in model_config["speaker_mapping"].items() | |
] | |
language_radio_choices = [ | |
(k, v) for k, v in model_config["language_mapping"].items() | |
] | |
use_default_emb_or_ref_radio_visible = False | |
if model_config["model"].tts_model.config.model_args.speaker_encoder_model_path: | |
use_default_emb_or_ref_radio_visible = True | |
return ( | |
gr.update( | |
choices=speaker_drop_down_choices, | |
value=speaker_drop_down_choices[0][1] | |
if len(speaker_drop_down_choices) > 0 | |
else None, | |
interactive=len(speaker_drop_down_choices) > 1, | |
), | |
gr.update( | |
choices=language_radio_choices, | |
value=language_radio_choices[0][1], | |
interactive=len(language_radio_choices) > 1, | |
), | |
gr.update(visible=use_default_emb_or_ref_radio_visible, value="預設語者"), | |
) | |
def use_default_emb_or_custom_radio_input(use_default_emb_or_custom): | |
if use_default_emb_or_custom == "客製化語者": | |
return gr.update(visible=True), gr.update(visible=False) | |
return gr.update(visible=False), gr.update(visible=True) | |
def language_radio_changed(language): | |
if language in g2p: | |
return gr.update(visible=False) | |
dialect_choices = [tag.split("_")[1] for tag in g2p.keys() if language in tag] | |
return gr.update( | |
choices=dialect_choices, | |
value=dialect_choices[0], | |
interactive=len(dialect_choices) > 1, | |
) | |
demo = gr.Blocks( | |
title="臺灣南島語語音合成系統", | |
css="@import url(https://tauhu.tw/tauhu-oo.css);", | |
theme=gr.themes.Default( | |
font=( | |
"tauhu-oo", | |
gr.themes.GoogleFont("Source Sans Pro"), | |
"ui-sans-serif", | |
"system-ui", | |
"sans-serif", | |
) | |
), | |
) | |
with demo: | |
default_model_id = list(models_config.keys())[0] | |
model_drop_down = gr.Dropdown( | |
models_config.keys(), | |
value=default_model_id, | |
label="模型", | |
) | |
use_default_emb_or_custom_radio = gr.Radio( | |
label="語者類型", | |
choices=["預設語者", "客製化語者"], | |
value="客製化語者", | |
visible=True, | |
show_label=False, | |
) | |
speaker_wav = gr.Audio( | |
label="客製化語音", | |
visible=True, | |
editable=False, | |
type="filepath", | |
waveform_options=gr.WaveformOptions( | |
show_controls=False, | |
sample_rate=16000, | |
), | |
) | |
speaker_drop_down = gr.Dropdown( | |
choices=[ | |
(k, v) | |
for k, v in models_config[default_model_id]["speaker_mapping"].items() | |
], | |
value=list(models_config[default_model_id]["speaker_mapping"].values())[0], | |
label="語者", | |
interactive=len(models_config[default_model_id]["speaker_mapping"]) > 1, | |
visible=False, | |
) | |
use_default_emb_or_custom_radio.change( | |
use_default_emb_or_custom_radio_input, | |
inputs=[use_default_emb_or_custom_radio], | |
outputs=[speaker_wav, speaker_drop_down], | |
) | |
default_language = list( | |
models_config[default_model_id]["language_mapping"].values() | |
)[0] | |
language_radio = gr.Radio( | |
choices=[ | |
(k, v) | |
for k, v in models_config[default_model_id]["language_mapping"].items() | |
], | |
value=default_language, | |
label="語言", | |
interactive=len(models_config[default_model_id]["language_mapping"]) > 1, | |
) | |
default_dialect_choices = [ | |
tag.split("_")[1] for tag in g2p.keys() if default_language in tag | |
] | |
dialect_radio = gr.Radio( | |
choices=default_dialect_choices, | |
value=default_dialect_choices[0], | |
label="方言", | |
interactive=len(default_dialect_choices) > 1, | |
) | |
language_radio.change( | |
language_radio_changed, inputs=[language_radio], outputs=[dialect_radio] | |
) | |
model_drop_down.input( | |
when_model_selected, | |
inputs=[model_drop_down], | |
outputs=[speaker_drop_down, language_radio, use_default_emb_or_custom_radio], | |
) | |
input_text = gr.Textbox( | |
label="輸入文字", | |
value="", | |
) | |
speed = gr.Slider(maximum=1.5, minimum=0.5, value=1, label="語速") | |
with open("DEMO.md") as tong: | |
gr.Markdown(tong.read()) | |
gr.Interface( | |
text_to_speech, | |
inputs=[ | |
model_drop_down, | |
use_default_emb_or_custom_radio, | |
speaker_wav, | |
speaker_drop_down, | |
language_radio, | |
dialect_radio, | |
speed, | |
input_text, | |
], | |
outputs=[ | |
gr.Audio(interactive=False, label="合成語音", show_download_button=True), | |
], | |
allow_flagging="auto", | |
) | |
gr.Examples( | |
[ | |
[ | |
"預設語者", | |
"formosan_dict_ami#wav/formosan_dict_ami/000002_2.31-6.09.wav", | |
"阿美", | |
"南勢", | |
"mikadavu ku vavainay, i vavahiyan, a luma’", | |
], | |
[ | |
"預設語者", | |
"formosan_dict_ami#wav/formosan_dict_ami/000035_0.00-3.69.wav", | |
"阿美", | |
"南勢", | |
"mikadavu ku vavainay, i vavahiyan, a luma’", | |
], | |
], | |
label="範例", | |
inputs=[ | |
use_default_emb_or_custom_radio, | |
speaker_drop_down, | |
language_radio, | |
dialect_radio, | |
input_text, | |
], | |
) | |
demo.launch() | |