Mahiruoshi
commited on
Commit
•
c3497ff
1
Parent(s):
b35751a
Update app.py
Browse files
app.py
CHANGED
@@ -365,6 +365,8 @@ def create_tts_fn(net_g,hps,speaker_id):
|
|
365 |
messages = []
|
366 |
messages.append({"role": "assistant", "content": text})
|
367 |
htm = to_html(messages)
|
|
|
|
|
368 |
if not extract:
|
369 |
t1 = time.time()
|
370 |
stn_tst = get_text(sle(language,text),hps)
|
@@ -491,11 +493,11 @@ if __name__ == '__main__':
|
|
491 |
btnVC = gr.Button("Send")
|
492 |
output1 = gr.Audio(label="采样率22050")
|
493 |
with gr.Accordion(label="Setting(TTS)", open=False):
|
494 |
-
input2 = gr.Dropdown(label="
|
495 |
input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale),以控制情感", value=0.6)
|
496 |
input5 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w),以控制音素长短", value=0.668)
|
497 |
input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
|
498 |
-
with gr.Accordion(label="Advanced Setting(GPT3.5
|
499 |
input3 = gr.Checkbox(value=False, label="长句切割(小说合成)")
|
500 |
inputxt = gr.File(label="Text")
|
501 |
btnbook = gr.Button("小说合成")
|
@@ -516,7 +518,7 @@ if __name__ == '__main__':
|
|
516 |
record_audio = gr.Audio(label="record your voice", source="microphone")
|
517 |
upload_audio = gr.Audio(label="or upload audio here", source="upload")
|
518 |
with gr.Accordion(label="方法2:由原说话人先进行tts后套娃,适用于合成中文等特殊场景", open=True):
|
519 |
-
text = gr.TextArea(label="Text", value='
|
520 |
language = gr.Dropdown(label="Language", choices=lan, value="自动", interactive=True)
|
521 |
n_scale = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale),以控制情感", value=0.6)
|
522 |
n_scale_w = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w),以控制音素长短", value=0.668)
|
@@ -534,6 +536,6 @@ if __name__ == '__main__':
|
|
534 |
"### <center> 请不要生成会对个人以及企划造成侵害的内容,自觉遵守相关法律,静止商业使用或让他人产生困扰\n"
|
535 |
"<div align='center'>从左到右分别是虹团,少歌中文特化版,以及五校混合版。这三个均为不同的模型,效果也有差异</div>\n"
|
536 |
"<div align='center'>因为我会时不时地更新模型,所以会碰到平台抽风问题,大部分情况下一天就能恢复了。</div>\n"
|
537 |
-
'<div align="center"><a
|
538 |
-
'<div align="center"><a>建议只在平台上体验最基础的功能,强烈建议将该仓库克隆至本地或者于colab
|
539 |
app.launch()
|
|
|
365 |
messages = []
|
366 |
messages.append({"role": "assistant", "content": text})
|
367 |
htm = to_html(messages)
|
368 |
+
if language == '自动':
|
369 |
+
l_scale = 1.1 if is_japanese(text) else 1.2
|
370 |
if not extract:
|
371 |
t1 = time.time()
|
372 |
stn_tst = get_text(sle(language,text),hps)
|
|
|
493 |
btnVC = gr.Button("Send")
|
494 |
output1 = gr.Audio(label="采样率22050")
|
495 |
with gr.Accordion(label="Setting(TTS)", open=False):
|
496 |
+
input2 = gr.Dropdown(label="参数及语言选择方式", choices=lan, value="自动", interactive=True)
|
497 |
input4 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale),以控制情感", value=0.6)
|
498 |
input5 = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w),以控制音素长短", value=0.668)
|
499 |
input6 = gr.Slider(minimum=0.1, maximum=10, label="duration", value=1)
|
500 |
+
with gr.Accordion(label="Advanced Setting(GPT3.5接口+小说合成,仅展示用,大部分功能用不了。需克隆本仓库后本地运行main.py)", open=False):
|
501 |
input3 = gr.Checkbox(value=False, label="长句切割(小说合成)")
|
502 |
inputxt = gr.File(label="Text")
|
503 |
btnbook = gr.Button("小说合成")
|
|
|
518 |
record_audio = gr.Audio(label="record your voice", source="microphone")
|
519 |
upload_audio = gr.Audio(label="or upload audio here", source="upload")
|
520 |
with gr.Accordion(label="方法2:由原说话人先进行tts后套娃,适用于合成中文等特殊场景", open=True):
|
521 |
+
text = gr.TextArea(label="Text", value='输入文本',lines = 1)
|
522 |
language = gr.Dropdown(label="Language", choices=lan, value="自动", interactive=True)
|
523 |
n_scale = gr.Slider(minimum=0, maximum=1.0, label="更改噪声比例(noise scale),以控制情感", value=0.6)
|
524 |
n_scale_w = gr.Slider(minimum=0, maximum=1.0, label="更改噪声偏差(noise scale w),以控制音素长短", value=0.668)
|
|
|
536 |
"### <center> 请不要生成会对个人以及企划造成侵害的内容,自觉遵守相关法律,静止商业使用或让他人产生困扰\n"
|
537 |
"<div align='center'>从左到右分别是虹团,少歌中文特化版,以及五校混合版。这三个均为不同的模型,效果也有差异</div>\n"
|
538 |
"<div align='center'>因为我会时不时地更新模型,所以会碰到平台抽风问题,大部分情况下一天就能恢复了。</div>\n"
|
539 |
+
'<div align="center"><a>参数说明:这个十分玄学,如果效果不佳可以将噪声比例和噪声偏差调节至0,这会完全随机化音频源。按照经验,合成日语时也可以将噪声比例调节至0.2-0.3区间,语调会正常一些。duration代表整体语速,可视情况调至1.1或1.2,目前已自动匹配,如需调整将language项调为日文或中文。</div>'
|
540 |
+
'<div align="center"><a>建议只在平台上体验最基础的功能,强烈建议将该仓库克隆至本地或者于colab运行,启动程序为main.py或app.py</div>')
|
541 |
app.launch()
|