blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-04-05 19:41:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

v3解锁并行推理;修复模型加载异步逻辑

Browse Source 
v3解锁并行推理;修复模型加载异步逻辑
This commit is contained in:
RVC-Boss 2025-04-01 16:29:52 +08:00 committed by GitHub
parent 28bdff356f
commit 6a60e5edb1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 22 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
GPT_SoVITS/inference_webui.py
View File

@ -213,6 +213,7 @@ def resample(audio_tensor, sr0):
#symbol_version-model_version-if_lora_v3 #symbol_version-model_version-if_lora_v3
from process_ckpt import get_sovits_version_from_path_fast,load_sovits_new from process_ckpt import get_sovits_version_from_path_fast,load_sovits_new
def change_sovits_weights(sovits_path,prompt_language=None,text_language=None): def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
yield [None]*10+[{"__type__": "update", "value":i18n("模型加载中，请等待"),"interactive":False}]
global vq_model, hps, version, model_version, dict_language,if_lora_v3 global vq_model, hps, version, model_version, dict_language,if_lora_v3
version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path) version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path)
# print(sovits_path,version, model_version, if_lora_v3) # print(sovits_path,version, model_version, if_lora_v3)
@ -238,7 +239,7 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
else: else:
visible_sample_steps=False visible_sample_steps=False
visible_inp_refs=True visible_inp_refs=True
yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False} yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False},None
dict_s2 = load_sovits_new(sovits_path) dict_s2 = load_sovits_new(sovits_path)
hps = dict_s2["config"] hps = dict_s2["config"]
@ -294,6 +295,7 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
# torch.save(vq_model.state_dict(),"merge_win.pth") # torch.save(vq_model.state_dict(),"merge_win.pth")
vq_model.eval() vq_model.eval()
yield [None]*10+[{"__type__": "update", "value":i18n("合成语音"),"interactive":True}]
with open("./weight.json")as f: with open("./weight.json")as f:
data=f.read() data=f.read()
data=json.loads(data) data=json.loads(data)
@ -877,7 +879,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
with gr.Row(): with gr.Row():
inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath", scale=13) inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath", scale=13)
with gr.Column(scale=13): with gr.Column(scale=13):
ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。")+i18n("v3暂不支持该模式，使用了会报错。"), value=False, interactive=True, show_label=True,scale=1) ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。")+i18n("v3暂不支持该模式，使用了会报错。"), value=False, interactive=True if model_version!="v3"else False, show_label=True,scale=1)
gr.Markdown(html_left(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。"))) gr.Markdown(html_left(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。")))
prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="", lines=5, max_lines=5,scale=1) prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="", lines=5, max_lines=5,scale=1)
with gr.Column(scale=14): with gr.Column(scale=14):
@ -915,7 +917,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
# phoneme=gr.Textbox(label=i18n("音素框"), value="") # phoneme=gr.Textbox(label=i18n("音素框"), value="")
# get_phoneme_button = gr.Button(i18n("目标文本转音素"), variant="primary") # get_phoneme_button = gr.Button(i18n("目标文本转音素"), variant="primary")
with gr.Row(): with gr.Row():
inference_button = gr.Button(i18n("合成语音"), variant="primary", size='lg', scale=25) inference_button = gr.Button(value=i18n("合成语音"), variant="primary", size='lg', scale=25)
output = gr.Audio(label=i18n("输出的语音"), scale=14) output = gr.Audio(label=i18n("输出的语音"), scale=14)
inference_button.click( inference_button.click(
@ -923,7 +925,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
[inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut, top_k, top_p, temperature, ref_text_free,speed,if_freeze,inp_refs,sample_steps,if_sr_Checkbox,pause_second_slider], [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut, top_k, top_p, temperature, ref_text_free,speed,if_freeze,inp_refs,sample_steps,if_sr_Checkbox,pause_second_slider],
[output], [output],
) )
SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,if_sr_Checkbox]) SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,if_sr_Checkbox,inference_button])
GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], []) GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], [])
# gr.Markdown(value=i18n("文本切分工具。太长的文本合成出来效果不一定好，所以太长建议先切。合成会根据文本的换行分开合成再拼起来。")) # gr.Markdown(value=i18n("文本切分工具。太长的文本合成出来效果不一定好，所以太长建议先切。合成会根据文本的换行分开合成再拼起来。"))

30
GPT_SoVITS/inference_webui_fast.py
View File

@ -41,12 +41,13 @@ gpt_path = os.environ.get("gpt_path", None)
sovits_path = os.environ.get("sovits_path", None) sovits_path = os.environ.get("sovits_path", None)
cnhubert_base_path = os.environ.get("cnhubert_base_path", None) cnhubert_base_path = os.environ.get("cnhubert_base_path", None)
bert_path = os.environ.get("bert_path", None) bert_path = os.environ.get("bert_path", None)
version=os.environ.get("version","v2") version=model_version=os.environ.get("version","v2")
import gradio as gr import gradio as gr
from TTS_infer_pack.TTS import TTS, TTS_Config, NO_PROMPT_ERROR from TTS_infer_pack.TTS import TTS, TTS_Config, NO_PROMPT_ERROR
from TTS_infer_pack.text_segmentation_method import get_method from TTS_infer_pack.text_segmentation_method import get_method
from tools.i18n.i18n import I18nAuto, scan_language_list from tools.i18n.i18n import I18nAuto, scan_language_list
from inference_webui import DictToAttrRecursive
language=os.environ.get("language","Auto") language=os.environ.get("language","Auto")
language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language
@ -221,19 +222,17 @@ def get_weights_names(GPT_weight_root, SoVITS_weight_root):
SoVITS_names, GPT_names = get_weights_names(GPT_weight_root, SoVITS_weight_root) SoVITS_names, GPT_names = get_weights_names(GPT_weight_root, SoVITS_weight_root)
from process_ckpt import get_sovits_version_from_path_fast from process_ckpt import get_sovits_version_from_path_fast,load_sovits_new
def change_sovits_weights(sovits_path,prompt_language=None,text_language=None): def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
global version, dict_language yield [None]*9+[{"__type__": "update", "value":i18n("模型加载中，请等待"),"interactive":False}]
global version, model_version, dict_language,if_lora_v3
version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path) version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path)
# print(sovits_path,version, model_version, if_lora_v3)
if if_lora_v3 and not os.path.exists(path_sovits_v3): if if_lora_v3==True and is_exist_s2gv3==False:
info= path_sovits_v3 + i18n("SoVITS V3 底模缺失，无法加载相应 LoRA 权重") info= "GPT_SoVITS/pretrained_models/s2Gv3.pth" + i18n("SoVITS V3 底模缺失，无法加载相应 LoRA 权重")
gr.Warning(info) gr.Warning(info)
raise FileExistsError(info) raise FileExistsError(info)
dict_language = dict_language_v1 if version =='v1' else dict_language_v2
tts_pipeline.init_vits_weights(sovits_path)
dict_language = dict_language_v1 if tts_pipeline.configs.version =='v1' else dict_language_v2
if prompt_language is not None and text_language is not None: if prompt_language is not None and text_language is not None:
if prompt_language in list(dict_language.keys()): if prompt_language in list(dict_language.keys()):
prompt_text_update, prompt_language_update = {'__type__':'update'}, {'__type__':'update', 'value':prompt_language} prompt_text_update, prompt_language_update = {'__type__':'update'}, {'__type__':'update', 'value':prompt_language}
@ -251,8 +250,11 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
else: else:
visible_sample_steps=False visible_sample_steps=False
visible_inp_refs=True visible_inp_refs=True
yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False} #prompt_language,text_language,prompt_text,prompt_language,text,text_language,inp_refs,ref_text_free,
yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "interactive": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "interactive": True if model_version!="v3"else False},None
tts_pipeline.init_vits_weights(sovits_path)
yield [None]*9+[{"__type__": "update", "value":i18n("合成语音"),"interactive":True}]
with open("./weight.json")as f: with open("./weight.json")as f:
data=f.read() data=f.read()
data=json.loads(data) data=json.loads(data)
@ -279,14 +281,14 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
gr.Markdown(value=i18n("*请上传并填写参考信息")) gr.Markdown(value=i18n("*请上传并填写参考信息"))
with gr.Row(): with gr.Row():
inp_ref = gr.Audio(label=i18n("主参考音频(请上传3~10秒内参考音频，超过会报错！)"), type="filepath") inp_ref = gr.Audio(label=i18n("主参考音频(请上传3~10秒内参考音频，超过会报错！)"), type="filepath")
inp_refs = gr.File(label=i18n("辅参考音频(可选多个，或不选)"),file_count="multiple") inp_refs = gr.File(label=i18n("辅参考音频(可选多个，或不选)"),file_count="multiple", visible=True if model_version!="v3"else False)
prompt_text = gr.Textbox(label=i18n("主参考音频的文本"), value="", lines=2) prompt_text = gr.Textbox(label=i18n("主参考音频的文本"), value="", lines=2)
with gr.Row(): with gr.Row():
prompt_language = gr.Dropdown( prompt_language = gr.Dropdown(
label=i18n("主参考音频的语种"), choices=list(dict_language.keys()), value=i18n("中文") label=i18n("主参考音频的语种"), choices=list(dict_language.keys()), value=i18n("中文")
) )
with gr.Column(): with gr.Column():
ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。"), value=False, interactive=True, show_label=True) ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。"), value=False, interactive=True if model_version!="v3"else False, show_label=True)
gr.Markdown(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。")) gr.Markdown(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。"))
with gr.Column(): with gr.Column():
@ -355,7 +357,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
[output, seed], [output, seed],
) )
stop_infer.click(tts_pipeline.stop, [], []) stop_infer.click(tts_pipeline.stop, [], [])
SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language]) SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,inference_button])#
GPT_dropdown.change(tts_pipeline.init_t2s_weights, [GPT_dropdown], []) GPT_dropdown.change(tts_pipeline.init_t2s_weights, [GPT_dropdown], [])
with gr.Group(): with gr.Group():