bacow/GPT-SoVITS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add DPO training

Browse Source
This commit is contained in:
liufenghua
2024-02-11 15:06:01 +08:00
parent 41041715a4
commit 070ac9b2b2
4 changed files with 176 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
GPT_SoVITS/inference_webui.py
View File

@@ -359,7 +359,7 @@ def merge_short_text_in_array(texts, threshold):
result[len(result) - 1] += text
return result
def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切")):
def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切"), top_k=20, top_p=0.6, temperature=0.6):
t0 = ttime()
prompt_language = dict_language[prompt_language]
text_language = dict_language[text_language]
@@ -438,7 +438,9 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
prompt,
bert,
# prompt_phone_len=ph_offset,
top_k=config["inference"]["top_k"],
top_k=top_k,
top_p=top_p,
temperature=temperature,
early_stop_num=hz * max_sec,
)
t3 = ttime()
@@ -615,6 +617,10 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
value=i18n("凑四句一切"),
interactive=True,
)
with gr.Row():
top_k = gr.Slider(minimum=1,maximum=100,step=1,label=i18n("top_k"),value=20,interactive=True)
top_p = gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("top_p"),value=0.6,interactive=True)
temperature = gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("temperature"),value=0.6,interactive=True)
inference_button = gr.Button(i18n("合成语音"), variant="primary")
output = gr.Audio(label=i18n("输出的语音"))