mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2026-01-09 20:07:02 +08:00
Merge d32d70c2b103e1d3e5b17a98d8311daa1db0ed88 into 165882d64f474b3563fa91adc1a679436ae9c3b8
This commit is contained in:
laubonghaudoi
GitHub
commit
5cdac67fc4
@ -1,6 +1,17 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
|
||||
import os
|
||||
import os.path
|
||||
import shutil
|
||||
import traceback
|
||||
from time import time as ttime
|
||||
|
||||
import torch
|
||||
from text.cleaner import clean_text
|
||||
from tqdm import tqdm
|
||||
from transformers import AutoModelForMaskedLM, AutoTokenizer
|
||||
|
||||
from tools.my_utils import clean_path
|
||||
|
||||
inp_text = os.environ.get("inp_text")
|
||||
inp_wav_dir = os.environ.get("inp_wav_dir")
|
||||
@ -11,17 +22,8 @@ if "_CUDA_VISIBLE_DEVICES" in os.environ:
|
||||
os.environ["CUDA_VISIBLE_DEVICES"] = os.environ["_CUDA_VISIBLE_DEVICES"]
|
||||
opt_dir = os.environ.get("opt_dir")
|
||||
bert_pretrained_dir = os.environ.get("bert_pretrained_dir")
|
||||
import torch
|
||||
is_half = eval(os.environ.get("is_half", "True")) and torch.cuda.is_available()
|
||||
version = os.environ.get('version', None)
|
||||
import sys, numpy as np, traceback, pdb
|
||||
import os.path
|
||||
from glob import glob
|
||||
from tqdm import tqdm
|
||||
from text.cleaner import clean_text
|
||||
from transformers import AutoModelForMaskedLM, AutoTokenizer
|
||||
import numpy as np
|
||||
from tools.my_utils import clean_path
|
||||
|
||||
# inp_text=sys.argv[1]
|
||||
# inp_wav_dir=sys.argv[2]
|
||||
@ -32,11 +34,8 @@ from tools.my_utils import clean_path
|
||||
# opt_dir="/data/docker/liujing04/gpt-vits/fine_tune_dataset/%s"%exp_name
|
||||
# bert_pretrained_dir="/data/docker/liujing04/bert-vits2/Bert-VITS2-master20231106/bert/chinese-roberta-wwm-ext-large"
|
||||
|
||||
from time import time as ttime
|
||||
import shutil
|
||||
|
||||
|
||||
def my_save(fea,path):#####fix issue: torch.save doesn't support chinese path
|
||||
def my_save(fea, path): # fix issue: torch.save doesn't support chinese path
|
||||
dir = os.path.dirname(path)
|
||||
name = os.path.basename(path)
|
||||
# tmp_path="%s/%s%s.pth"%(dir,ttime(),i_part)
|
||||
@ -56,11 +55,13 @@ if os.path.exists(txt_path) == False:
|
||||
# device = "mps"
|
||||
else:
|
||||
device = "cpu"
|
||||
if os.path.exists(bert_pretrained_dir):...
|
||||
else:raise FileNotFoundError(bert_pretrained_dir)
|
||||
if os.path.exists(bert_pretrained_dir):
|
||||
...
|
||||
else:
|
||||
raise FileNotFoundError(bert_pretrained_dir)
|
||||
tokenizer = AutoTokenizer.from_pretrained(bert_pretrained_dir)
|
||||
bert_model = AutoModelForMaskedLM.from_pretrained(bert_pretrained_dir)
|
||||
if is_half == True:
|
||||
if is_half:
|
||||
bert_model = bert_model.half().to(device)
|
||||
else:
|
||||
bert_model = bert_model.to(device)
|
||||
@ -126,7 +127,7 @@ if os.path.exists(txt_path) == False:
|
||||
"YUE": "yue",
|
||||
"Yue": "yue",
|
||||
}
|
||||
for line in lines[int(i_part) :: int(all_parts)]:
|
||||
for line in tqdm(lines[int(i_part):: int(all_parts)]):
|
||||
try:
|
||||
wav_name, spk_name, language, text = line.split("|")
|
||||
# todo.append([name,text,"zh"])
|
||||
|
||||
@ -1,6 +1,23 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
Step 2 of data preparation: Extract HuBERT features from the audio files,
|
||||
and resample the audio to 32kHz and saving it.
|
||||
"""
|
||||
|
||||
import os
|
||||
import shutil
|
||||
import sys
|
||||
import traceback
|
||||
from time import time as ttime
|
||||
|
||||
import librosa
|
||||
import numpy as np
|
||||
import torch
|
||||
from feature_extractor import cnhubert
|
||||
from scipy.io import wavfile
|
||||
from tqdm import tqdm
|
||||
|
||||
from tools.my_utils import clean_path, load_audio
|
||||
|
||||
import sys,os
|
||||
inp_text = os.environ.get("inp_text")
|
||||
inp_wav_dir = os.environ.get("inp_wav_dir")
|
||||
exp_name = os.environ.get("exp_name")
|
||||
@ -8,18 +25,16 @@ i_part= os.environ.get("i_part")
|
||||
all_parts = os.environ.get("all_parts")
|
||||
if "_CUDA_VISIBLE_DEVICES" in os.environ:
|
||||
os.environ["CUDA_VISIBLE_DEVICES"] = os.environ["_CUDA_VISIBLE_DEVICES"]
|
||||
from feature_extractor import cnhubert
|
||||
|
||||
opt_dir = os.environ.get("opt_dir")
|
||||
cnhubert.cnhubert_base_path = os.environ.get("cnhubert_base_dir")
|
||||
import torch
|
||||
|
||||
is_half = eval(os.environ.get("is_half", "True")) and torch.cuda.is_available()
|
||||
|
||||
import pdb,traceback,numpy as np,logging
|
||||
from scipy.io import wavfile
|
||||
import librosa
|
||||
|
||||
now_dir = os.getcwd()
|
||||
sys.path.append(now_dir)
|
||||
from tools.my_utils import load_audio,clean_path
|
||||
|
||||
|
||||
# from config import cnhubert_base_path
|
||||
# cnhubert.cnhubert_base_path=cnhubert_base_path
|
||||
@ -32,15 +47,14 @@ from tools.my_utils import load_audio,clean_path
|
||||
# cnhubert.cnhubert_base_path=sys.argv[7]
|
||||
# opt_dir="/data/docker/liujing04/gpt-vits/fine_tune_dataset/%s"%exp_name
|
||||
|
||||
from time import time as ttime
|
||||
import shutil
|
||||
def my_save(fea,path):#####fix issue: torch.save doesn't support chinese path
|
||||
def my_save(fea, path): # fix issue: torch.save doesn't support chinese path
|
||||
dir = os.path.dirname(path)
|
||||
name = os.path.basename(path)
|
||||
# tmp_path="%s/%s%s.pth"%(dir,ttime(),i_part)
|
||||
tmp_path="%s%s.pth"%(ttime(),i_part)
|
||||
tmp_path = f"{ttime()}{i_part}.pth"
|
||||
torch.save(fea, tmp_path)
|
||||
shutil.move(tmp_path,"%s/%s"%(dir,name))
|
||||
shutil.move(tmp_path, f"{dir}/{name}")
|
||||
|
||||
|
||||
hubert_dir = "%s/4-cnhubert" % (opt_dir)
|
||||
wav32dir = "%s/5-wav32k" % (opt_dir)
|
||||
@ -56,6 +70,8 @@ if torch.cuda.is_available():
|
||||
# device = "mps"
|
||||
else:
|
||||
device = "cpu"
|
||||
|
||||
|
||||
model = cnhubert.get_model()
|
||||
# is_half=False
|
||||
if (is_half == True):
|
||||
@ -64,47 +80,66 @@ else:
|
||||
model = model.to(device)
|
||||
|
||||
nan_fails = []
|
||||
|
||||
|
||||
def name2go(wav_name, wav_path):
|
||||
"""
|
||||
Extract HuBERT features from the audio files, and resample the audio to 32kHz and saving it.
|
||||
"""
|
||||
# Skip if the file already exists
|
||||
hubert_path = "%s/%s.pt" % (hubert_dir, wav_name)
|
||||
if(os.path.exists(hubert_path)):return
|
||||
tmp_audio = load_audio(wav_path, 32000)
|
||||
tmp_max = np.abs(tmp_audio).max()
|
||||
if tmp_max > 2.2:
|
||||
print("%s-filtered,%s" % (wav_name, tmp_max))
|
||||
if (os.path.exists(hubert_path)):
|
||||
return
|
||||
|
||||
# Load the audio file in 32kHz sampling rate
|
||||
tmp_audio = load_audio(wav_path, 32000)
|
||||
|
||||
# Check the maximum amplitude of the audio file
|
||||
tmp_max = np.abs(tmp_audio).max()
|
||||
# Skip if the maximum amplitude is too high (volume is too loud)
|
||||
if tmp_max > 2.2:
|
||||
print(f"{wav_name}-filtered,{tmp_max}")
|
||||
return
|
||||
# Normalize the audio
|
||||
tmp_audio32 = (tmp_audio / tmp_max * (maxx * alpha * 32768)) + ((1 - alpha) * 32768) * tmp_audio
|
||||
tmp_audio32b = (tmp_audio / tmp_max * (maxx * alpha * 1145.14)) + ((1 - alpha) * 1145.14) * tmp_audio
|
||||
tmp_audio = librosa.resample(
|
||||
tmp_audio32b, orig_sr=32000, target_sr=16000
|
||||
) # 不是重采样问题
|
||||
tensor_wav16 = torch.from_numpy(tmp_audio)
|
||||
if (is_half == True):
|
||||
|
||||
# if half-precision is enabled, convert the tensor to half-precision
|
||||
if is_half:
|
||||
tensor_wav16 = tensor_wav16.half().to(device)
|
||||
else:
|
||||
tensor_wav16 = tensor_wav16.to(device)
|
||||
|
||||
# Extract HuBERT features from the audio file
|
||||
ssl = model.model(tensor_wav16.unsqueeze(0))["last_hidden_state"].transpose(1, 2).cpu() # torch.Size([1, 768, 215])
|
||||
|
||||
if np.isnan(ssl.detach().numpy()).sum() != 0:
|
||||
nan_fails.append((wav_name, wav_path))
|
||||
print("nan filtered:%s"%wav_name)
|
||||
print(f"nan filtered:{wav_name}")
|
||||
return
|
||||
wavfile.write(
|
||||
"%s/%s"%(wav32dir,wav_name),
|
||||
f"{wav32dir}/{wav_name}",
|
||||
32000,
|
||||
tmp_audio32.astype("int16"),
|
||||
)
|
||||
my_save(ssl, hubert_path)
|
||||
|
||||
|
||||
with open(inp_text, "r", encoding="utf8")as f:
|
||||
lines = f.read().strip("\n").split("\n")
|
||||
|
||||
for line in lines[int(i_part)::int(all_parts)]:
|
||||
for line in tqdm(lines[int(i_part)::int(all_parts)]):
|
||||
try:
|
||||
# wav_name,text=line.split("\t")
|
||||
wav_name, spk_name, language, text = line.split("|")
|
||||
wav_name = clean_path(wav_name)
|
||||
if (inp_wav_dir != "" and inp_wav_dir != None):
|
||||
if (inp_wav_dir != "" and inp_wav_dir is not None):
|
||||
wav_name = os.path.basename(wav_name)
|
||||
wav_path = "%s/%s"%(inp_wav_dir, wav_name)
|
||||
wav_path = f"{inp_wav_dir}/{wav_name}"
|
||||
|
||||
else:
|
||||
wav_path = wav_name
|
||||
@ -113,7 +148,7 @@ for line in lines[int(i_part)::int(all_parts)]:
|
||||
except:
|
||||
print(line, traceback.format_exc())
|
||||
|
||||
if(len(nan_fails)>0 and is_half==True):
|
||||
if (len(nan_fails) > 0 and is_half):
|
||||
is_half = False
|
||||
model = model.float()
|
||||
for wav in nan_fails:
|
||||
|
||||
@ -1,4 +1,12 @@
|
||||
import logging
|
||||
import os
|
||||
import sys
|
||||
import traceback
|
||||
|
||||
import torch
|
||||
import utils
|
||||
|
||||
from tools.my_utils import clean_path
|
||||
|
||||
inp_text = os.environ.get("inp_text")
|
||||
exp_name = os.environ.get("exp_name")
|
||||
@ -10,8 +18,10 @@ opt_dir = os.environ.get("opt_dir")
|
||||
pretrained_s2G = os.environ.get("pretrained_s2G")
|
||||
s2config_path = os.environ.get("s2config_path")
|
||||
|
||||
if os.path.exists(pretrained_s2G):...
|
||||
else:raise FileNotFoundError(pretrained_s2G)
|
||||
if os.path.exists(pretrained_s2G):
|
||||
...
|
||||
else:
|
||||
raise FileNotFoundError(pretrained_s2G)
|
||||
# version=os.environ.get("version","v2")
|
||||
size = os.path.getsize(pretrained_s2G)
|
||||
if size < 82978 * 1024:
|
||||
@ -24,24 +34,14 @@ elif size < 700 * 1024 * 1024:
|
||||
version = "v2"
|
||||
else:
|
||||
version = "v3"
|
||||
import torch
|
||||
is_half = eval(os.environ.get("is_half", "True")) and torch.cuda.is_available()
|
||||
import math, traceback
|
||||
import multiprocessing
|
||||
import sys, pdb
|
||||
|
||||
now_dir = os.getcwd()
|
||||
sys.path.append(now_dir)
|
||||
from random import shuffle
|
||||
import torch.multiprocessing as mp
|
||||
from glob import glob
|
||||
from tqdm import tqdm
|
||||
import logging, librosa, utils
|
||||
if version != "v3":
|
||||
from module.models import SynthesizerTrn
|
||||
else:
|
||||
from module.models import SynthesizerTrnV3 as SynthesizerTrn
|
||||
from tools.my_utils import clean_path
|
||||
logging.getLogger("numba").setLevel(logging.WARNING)
|
||||
# from config import pretrained_s2G
|
||||
|
||||
|
||||
209
webui.py
209
webui.py
@ -1,15 +1,45 @@
|
||||
import os,sys
|
||||
if len(sys.argv)==1:sys.argv.append('v2')
|
||||
import json
|
||||
import os
|
||||
import platform
|
||||
import re
|
||||
import shutil
|
||||
import signal
|
||||
import site
|
||||
import subprocess
|
||||
import sys
|
||||
import traceback
|
||||
import warnings
|
||||
from multiprocessing import cpu_count
|
||||
from subprocess import Popen
|
||||
|
||||
import gradio as gr
|
||||
import psutil
|
||||
import torch
|
||||
import yaml
|
||||
|
||||
from config import (
|
||||
exp_root,
|
||||
infer_device,
|
||||
is_half,
|
||||
is_share,
|
||||
python_exec,
|
||||
webui_port_infer_tts,
|
||||
webui_port_main,
|
||||
webui_port_subfix,
|
||||
webui_port_uvr5,
|
||||
)
|
||||
from tools import my_utils
|
||||
from tools.asr.config import asr_dict
|
||||
from tools.i18n.i18n import I18nAuto, scan_language_list
|
||||
from tools.my_utils import check_details, check_for_existance
|
||||
|
||||
if len(sys.argv) == 1:
|
||||
sys.argv.append('v2')
|
||||
version = "v1"if sys.argv[1] == "v1" else "v2"
|
||||
os.environ["version"] = version
|
||||
now_dir = os.getcwd()
|
||||
sys.path.insert(0, now_dir)
|
||||
import warnings
|
||||
warnings.filterwarnings("ignore")
|
||||
import json,yaml,torch,pdb,re,shutil
|
||||
import platform
|
||||
import psutil
|
||||
import signal
|
||||
os.environ['TORCH_DISTRIBUTED_DEBUG'] = 'INFO'
|
||||
torch.manual_seed(233333)
|
||||
tmp = os.path.join(now_dir, "TEMP")
|
||||
@ -17,7 +47,8 @@ os.makedirs(tmp, exist_ok=True)
|
||||
os.environ["TEMP"] = tmp
|
||||
if (os.path.exists(tmp)):
|
||||
for name in os.listdir(tmp):
|
||||
if(name=="jieba.cache"):continue
|
||||
if (name == "jieba.cache"):
|
||||
continue
|
||||
path = "%s/%s" % (tmp, name)
|
||||
delete = os.remove if os.path.isfile(path) else shutil.rmtree
|
||||
try:
|
||||
@ -25,13 +56,12 @@ if(os.path.exists(tmp)):
|
||||
except Exception as e:
|
||||
print(str(e))
|
||||
pass
|
||||
import site
|
||||
import traceback
|
||||
site_packages_roots = []
|
||||
for path in site.getsitepackages():
|
||||
if "packages" in path:
|
||||
site_packages_roots.append(path)
|
||||
if(site_packages_roots==[]):site_packages_roots=["%s/runtime/Lib/site-packages" % now_dir]
|
||||
if (site_packages_roots == []):
|
||||
site_packages_roots = ["%s/runtime/Lib/site-packages" % now_dir]
|
||||
# os.environ["OPENBLAS_NUM_THREADS"] = "4"
|
||||
os.environ["no_proxy"] = "localhost, 127.0.0.1, ::1"
|
||||
os.environ["all_proxy"] = ""
|
||||
@ -45,28 +75,17 @@ for site_packages_root in site_packages_roots:
|
||||
% (now_dir, now_dir, now_dir, now_dir, now_dir, now_dir)
|
||||
)
|
||||
break
|
||||
except PermissionError as e:
|
||||
except PermissionError:
|
||||
traceback.print_exc()
|
||||
from tools import my_utils
|
||||
import shutil
|
||||
import pdb
|
||||
import subprocess
|
||||
from subprocess import Popen
|
||||
import signal
|
||||
from config import python_exec,infer_device,is_half,exp_root,webui_port_main,webui_port_infer_tts,webui_port_uvr5,webui_port_subfix,is_share
|
||||
from tools.i18n.i18n import I18nAuto, scan_language_list
|
||||
language = sys.argv[-1] if sys.argv[-1] in scan_language_list() else "Auto"
|
||||
os.environ["language"] = language
|
||||
i18n = I18nAuto(language=language)
|
||||
from scipy.io import wavfile
|
||||
from tools.my_utils import load_audio, check_for_existance, check_details
|
||||
from multiprocessing import cpu_count
|
||||
# os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1' # 当遇到mps不支持的步骤时使用cpu
|
||||
try:
|
||||
import gradio.analytics as analytics
|
||||
analytics.version_check = lambda: None
|
||||
except:...
|
||||
import gradio as gr
|
||||
except:
|
||||
...
|
||||
n_cpu = cpu_count()
|
||||
|
||||
ngpu = torch.cuda.device_count()
|
||||
@ -92,6 +111,7 @@ if torch.cuda.is_available() or ngpu != 0:
|
||||
# gpu_infos.append("%s\t%s" % ("0", "Apple GPU"))
|
||||
# mem.append(psutil.virtual_memory().total/ 1024 / 1024 / 1024) # 实测使用系统内存作为显存不会爆显存
|
||||
|
||||
|
||||
def set_default():
|
||||
global default_batch_size, default_max_batch_size, gpu_info, default_sovits_epoch, default_sovits_save_every_epoch, max_sovits_epoch, max_sovits_save_every_epoch, default_batch_size_s1, if_force_ckpt
|
||||
if_force_ckpt = False
|
||||
@ -139,23 +159,32 @@ def set_default():
|
||||
default_batch_size_s1 = max(1, default_batch_size_s1)
|
||||
default_max_batch_size = default_batch_size * 3
|
||||
|
||||
|
||||
set_default()
|
||||
|
||||
gpus = "-".join([i[0] for i in gpu_infos])
|
||||
default_gpu_numbers = str(sorted(list(set_gpu_numbers))[0])
|
||||
|
||||
|
||||
def fix_gpu_number(input): # 将越界的number强制改到界内
|
||||
try:
|
||||
if(int(input)not in set_gpu_numbers):return default_gpu_numbers
|
||||
except:return input
|
||||
if (int(input)not in set_gpu_numbers):
|
||||
return default_gpu_numbers
|
||||
except:
|
||||
return input
|
||||
return input
|
||||
|
||||
|
||||
def fix_gpu_numbers(inputs):
|
||||
output = []
|
||||
try:
|
||||
for input in inputs.split(","):output.append(str(fix_gpu_number(input)))
|
||||
for input in inputs.split(","):
|
||||
output.append(str(fix_gpu_number(input)))
|
||||
return ",".join(output)
|
||||
except:
|
||||
return inputs
|
||||
|
||||
|
||||
pretrained_sovits_name = ["GPT_SoVITS/pretrained_models/s2G488k.pth", "GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s2G2333k.pth", "GPT_SoVITS/pretrained_models/s2Gv3.pth"]
|
||||
pretrained_gpt_name = ["GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt", "GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt", "GPT_SoVITS/pretrained_models/s1v3.ckpt"]
|
||||
|
||||
@ -170,31 +199,41 @@ if _:
|
||||
|
||||
_ = [[], []]
|
||||
for i in range(3):
|
||||
if os.path.exists(pretrained_gpt_name[i]):_[0].append(pretrained_gpt_name[i])
|
||||
else:_[0].append("")##没有下pretrained模型的，说不定他们是想自己从零训底模呢
|
||||
if os.path.exists(pretrained_sovits_name[i]):_[-1].append(pretrained_sovits_name[i])
|
||||
else:_[-1].append("")
|
||||
if os.path.exists(pretrained_gpt_name[i]):
|
||||
_[0].append(pretrained_gpt_name[i])
|
||||
else:
|
||||
_[0].append("") # 没有下pretrained模型的，说不定他们是想自己从零训底模呢
|
||||
if os.path.exists(pretrained_sovits_name[i]):
|
||||
_[-1].append(pretrained_sovits_name[i])
|
||||
else:
|
||||
_[-1].append("")
|
||||
pretrained_gpt_name, pretrained_sovits_name = _
|
||||
|
||||
SoVITS_weight_root = ["SoVITS_weights", "SoVITS_weights_v2", "SoVITS_weights_v3"]
|
||||
GPT_weight_root = ["GPT_weights", "GPT_weights_v2", "GPT_weights_v3"]
|
||||
for root in SoVITS_weight_root + GPT_weight_root:
|
||||
os.makedirs(root, exist_ok=True)
|
||||
|
||||
|
||||
def get_weights_names():
|
||||
SoVITS_names = [name for name in pretrained_sovits_name if name != ""]
|
||||
for path in SoVITS_weight_root:
|
||||
for name in os.listdir(path):
|
||||
if name.endswith(".pth"): SoVITS_names.append("%s/%s" % (path, name))
|
||||
if name.endswith(".pth"):
|
||||
SoVITS_names.append("%s/%s" % (path, name))
|
||||
GPT_names = [name for name in pretrained_gpt_name if name != ""]
|
||||
for path in GPT_weight_root:
|
||||
for name in os.listdir(path):
|
||||
if name.endswith(".ckpt"): GPT_names.append("%s/%s" % (path, name))
|
||||
if name.endswith(".ckpt"):
|
||||
GPT_names.append("%s/%s" % (path, name))
|
||||
return SoVITS_names, GPT_names
|
||||
|
||||
|
||||
SoVITS_names, GPT_names = get_weights_names()
|
||||
for path in SoVITS_weight_root + GPT_weight_root:
|
||||
os.makedirs(path, exist_ok=True)
|
||||
|
||||
|
||||
def custom_sort_key(s):
|
||||
# 使用正则表达式提取字符串中的数字部分和非数字部分
|
||||
parts = re.split('(\d+)', s)
|
||||
@ -202,16 +241,19 @@ def custom_sort_key(s):
|
||||
parts = [int(part) if part.isdigit() else part for part in parts]
|
||||
return parts
|
||||
|
||||
|
||||
def change_choices():
|
||||
SoVITS_names, GPT_names = get_weights_names()
|
||||
return {"choices": sorted(SoVITS_names, key=custom_sort_key), "__type__": "update"}, {"choices": sorted(GPT_names, key=custom_sort_key), "__type__": "update"}
|
||||
|
||||
|
||||
p_label = None
|
||||
p_uvr5 = None
|
||||
p_asr = None
|
||||
p_denoise = None
|
||||
p_tts_inference = None
|
||||
|
||||
|
||||
def kill_proc_tree(pid, including_parent=True):
|
||||
try:
|
||||
parent = psutil.Process(pid)
|
||||
@ -231,7 +273,10 @@ def kill_proc_tree(pid, including_parent=True):
|
||||
except OSError:
|
||||
pass
|
||||
|
||||
|
||||
system = platform.system()
|
||||
|
||||
|
||||
def kill_process(pid, process_name=""):
|
||||
if (system == "Windows"):
|
||||
cmd = "taskkill /t /f /pid %s" % pid
|
||||
@ -241,6 +286,7 @@ def kill_process(pid, process_name=""):
|
||||
kill_proc_tree(pid)
|
||||
print(process_name + i18n("进程已终止"))
|
||||
|
||||
|
||||
def process_info(process_name="", indicator=""):
|
||||
if indicator == "opened":
|
||||
return process_name + i18n("已开启")
|
||||
@ -263,7 +309,10 @@ def process_info(process_name="", indicator=""):
|
||||
else:
|
||||
return process_name
|
||||
|
||||
|
||||
process_name_subfix = i18n("音频标注WebUI")
|
||||
|
||||
|
||||
def change_label(path_list):
|
||||
global p_label
|
||||
if p_label is None:
|
||||
@ -278,7 +327,10 @@ def change_label(path_list):
|
||||
p_label = None
|
||||
yield process_info(process_name_subfix, "closed"), {'__type__': 'update', 'visible': True}, {'__type__': 'update', 'visible': False}
|
||||
|
||||
|
||||
process_name_uvr5 = i18n("人声分离WebUI")
|
||||
|
||||
|
||||
def change_uvr5():
|
||||
global p_uvr5
|
||||
if p_uvr5 is None:
|
||||
@ -291,14 +343,17 @@ def change_uvr5():
|
||||
p_uvr5 = None
|
||||
yield process_info(process_name_uvr5, "closed"), {'__type__': 'update', 'visible': True}, {'__type__': 'update', 'visible': False}
|
||||
|
||||
|
||||
process_name_tts = i18n("TTS推理WebUI")
|
||||
|
||||
|
||||
def change_tts_inference(bert_path, cnhubert_base_path, gpu_number, gpt_path, sovits_path, batched_infer_enabled):
|
||||
global p_tts_inference
|
||||
if batched_infer_enabled:
|
||||
cmd = '"%s" GPT_SoVITS/inference_webui_fast.py "%s"' % (python_exec, language)
|
||||
else:
|
||||
cmd = '"%s" GPT_SoVITS/inference_webui.py "%s"' % (python_exec, language)
|
||||
#####v3暂不支持加速推理
|
||||
# v3暂不支持加速推理
|
||||
if version == "v3":
|
||||
cmd = '"%s" GPT_SoVITS/inference_webui.py "%s"' % (python_exec, language)
|
||||
if p_tts_inference is None:
|
||||
@ -318,9 +373,10 @@ def change_tts_inference(bert_path,cnhubert_base_path,gpu_number,gpt_path,sovits
|
||||
p_tts_inference = None
|
||||
yield process_info(process_name_tts, "closed"), {'__type__': 'update', 'visible': True}, {'__type__': 'update', 'visible': False}
|
||||
|
||||
from tools.asr.config import asr_dict
|
||||
|
||||
process_name_asr = i18n("语音识别")
|
||||
|
||||
|
||||
def open_asr(asr_inp_dir, asr_opt_dir, asr_model, asr_model_size, asr_lang, asr_precision):
|
||||
global p_asr
|
||||
if p_asr is None:
|
||||
@ -345,6 +401,7 @@ def open_asr(asr_inp_dir, asr_opt_dir, asr_model, asr_model_size, asr_lang, asr_
|
||||
else:
|
||||
yield process_info(process_name_asr, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}, {"__type__": "update"}, {"__type__": "update"}, {"__type__": "update"}
|
||||
|
||||
|
||||
def close_asr():
|
||||
global p_asr
|
||||
if p_asr is not None:
|
||||
@ -352,7 +409,10 @@ def close_asr():
|
||||
p_asr = None
|
||||
return process_info(process_name_asr, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
process_name_denoise = i18n("语音降噪")
|
||||
|
||||
|
||||
def open_denoise(denoise_inp_dir, denoise_opt_dir):
|
||||
global p_denoise
|
||||
if (p_denoise == None):
|
||||
@ -370,6 +430,7 @@ def open_denoise(denoise_inp_dir, denoise_opt_dir):
|
||||
else:
|
||||
yield process_info(process_name_denoise, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}, {"__type__": "update"}, {"__type__": "update"}
|
||||
|
||||
|
||||
def close_denoise():
|
||||
global p_denoise
|
||||
if p_denoise is not None:
|
||||
@ -377,8 +438,11 @@ def close_denoise():
|
||||
p_denoise = None
|
||||
return process_info(process_name_denoise, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
p_train_SoVITS = None
|
||||
process_name_sovits = i18n("SoVITS训练")
|
||||
|
||||
|
||||
def open1Ba(batch_size, total_epoch, exp_name, text_low_lr_rate, if_save_latest, if_save_every_weights, save_every_epoch, gpu_numbers1Ba, pretrained_s2G, pretrained_s2D, if_grad_ckpt, lora_rank):
|
||||
global p_train_SoVITS
|
||||
if (p_train_SoVITS == None):
|
||||
@ -409,7 +473,8 @@ def open1Ba(batch_size,total_epoch,exp_name,text_low_lr_rate,if_save_latest,if_s
|
||||
data["name"] = exp_name
|
||||
data["version"] = version
|
||||
tmp_config_path = "%s/tmp_s2.json" % tmp
|
||||
with open(tmp_config_path,"w")as f:f.write(json.dumps(data))
|
||||
with open(tmp_config_path, "w")as f:
|
||||
f.write(json.dumps(data))
|
||||
if version in ["v1", "v2"]:
|
||||
cmd = '"%s" GPT_SoVITS/s2_train.py --config "%s"' % (python_exec, tmp_config_path)
|
||||
else:
|
||||
@ -423,6 +488,7 @@ def open1Ba(batch_size,total_epoch,exp_name,text_low_lr_rate,if_save_latest,if_s
|
||||
else:
|
||||
yield process_info(process_name_sovits, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1Ba():
|
||||
global p_train_SoVITS
|
||||
if p_train_SoVITS is not None:
|
||||
@ -430,8 +496,11 @@ def close1Ba():
|
||||
p_train_SoVITS = None
|
||||
return process_info(process_name_sovits, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
p_train_GPT = None
|
||||
process_name_gpt = i18n("GPT训练")
|
||||
|
||||
|
||||
def open1Bb(batch_size, total_epoch, exp_name, if_dpo, if_save_latest, if_save_every_weights, save_every_epoch, gpu_numbers, pretrained_s1):
|
||||
global p_train_GPT
|
||||
if (p_train_GPT == None):
|
||||
@ -462,7 +531,8 @@ def open1Bb(batch_size,total_epoch,exp_name,if_dpo,if_save_latest,if_save_every_
|
||||
os.environ["_CUDA_VISIBLE_DEVICES"] = fix_gpu_numbers(gpu_numbers.replace("-", ","))
|
||||
os.environ["hz"] = "25hz"
|
||||
tmp_config_path = "%s/tmp_s1.yaml" % tmp
|
||||
with open(tmp_config_path, "w") as f:f.write(yaml.dump(data, default_flow_style=False))
|
||||
with open(tmp_config_path, "w") as f:
|
||||
f.write(yaml.dump(data, default_flow_style=False))
|
||||
# cmd = '"%s" GPT_SoVITS/s1_train.py --config_file "%s" --train_semantic_path "%s/6-name2semantic.tsv" --train_phoneme_path "%s/2-name2text.txt" --output_dir "%s/logs_s1"'%(python_exec,tmp_config_path,s1_dir,s1_dir,s1_dir)
|
||||
cmd = '"%s" GPT_SoVITS/s1_train.py --config_file "%s" ' % (python_exec, tmp_config_path)
|
||||
yield process_info(process_name_gpt, "opened"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
@ -474,6 +544,7 @@ def open1Bb(batch_size,total_epoch,exp_name,if_dpo,if_save_latest,if_save_every_
|
||||
else:
|
||||
yield process_info(process_name_gpt, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1Bb():
|
||||
global p_train_GPT
|
||||
if p_train_GPT is not None:
|
||||
@ -481,8 +552,11 @@ def close1Bb():
|
||||
p_train_GPT = None
|
||||
return process_info(process_name_gpt, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
ps_slice = []
|
||||
process_name_slice = i18n("语音切分")
|
||||
|
||||
|
||||
def open_slice(inp, opt_root, threshold, min_length, min_interval, hop_size, max_sil_kept, _max, alpha, n_parts):
|
||||
global ps_slice
|
||||
inp = my_utils.clean_path(inp)
|
||||
@ -491,8 +565,10 @@ def open_slice(inp,opt_root,threshold,min_length,min_interval,hop_size,max_sil_k
|
||||
if (os.path.exists(inp) == False):
|
||||
yield i18n("输入路径不存在"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}, {"__type__": "update"}, {"__type__": "update"}, {"__type__": "update"}
|
||||
return
|
||||
if os.path.isfile(inp):n_parts=1
|
||||
elif os.path.isdir(inp):pass
|
||||
if os.path.isfile(inp):
|
||||
n_parts = 1
|
||||
elif os.path.isdir(inp):
|
||||
pass
|
||||
else:
|
||||
yield i18n("输入路径存在但不可用"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}, {"__type__": "update"}, {"__type__": "update"}, {"__type__": "update"}
|
||||
return
|
||||
@ -510,6 +586,7 @@ def open_slice(inp,opt_root,threshold,min_length,min_interval,hop_size,max_sil_k
|
||||
else:
|
||||
yield process_info(process_name_slice, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}, {"__type__": "update"}, {"__type__": "update"}, {"__type__": "update"}
|
||||
|
||||
|
||||
def close_slice():
|
||||
global ps_slice
|
||||
if (ps_slice != []):
|
||||
@ -521,8 +598,11 @@ def close_slice():
|
||||
ps_slice = []
|
||||
return process_info(process_name_slice, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
ps1a = []
|
||||
process_name_1a = i18n("文本分词与特征提取")
|
||||
|
||||
|
||||
def open1a(inp_text, inp_wav_dir, exp_name, gpu_numbers, bert_pretrained_dir):
|
||||
global ps1a
|
||||
inp_text = my_utils.clean_path(inp_text)
|
||||
@ -574,6 +654,7 @@ def open1a(inp_text,inp_wav_dir,exp_name,gpu_numbers,bert_pretrained_dir):
|
||||
else:
|
||||
yield process_info(process_name_1a, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1a():
|
||||
global ps1a
|
||||
if ps1a != []:
|
||||
@ -585,8 +666,11 @@ def close1a():
|
||||
ps1a = []
|
||||
return process_info(process_name_1a, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
ps1b = []
|
||||
process_name_1b = i18n("语音自监督特征提取")
|
||||
|
||||
|
||||
def open1b(inp_text, inp_wav_dir, exp_name, gpu_numbers, ssl_pretrained_dir):
|
||||
global ps1b
|
||||
inp_text = my_utils.clean_path(inp_text)
|
||||
@ -625,6 +709,7 @@ def open1b(inp_text,inp_wav_dir,exp_name,gpu_numbers,ssl_pretrained_dir):
|
||||
else:
|
||||
yield process_info(process_name_1b, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1b():
|
||||
global ps1b
|
||||
if (ps1b != []):
|
||||
@ -636,8 +721,11 @@ def close1b():
|
||||
ps1b = []
|
||||
return process_info(process_name_1b, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
ps1c = []
|
||||
process_name_1c = i18n("语义Token提取")
|
||||
|
||||
|
||||
def open1c(inp_text, exp_name, gpu_numbers, pretrained_s2G_path):
|
||||
global ps1c
|
||||
inp_text = my_utils.clean_path(inp_text)
|
||||
@ -685,6 +773,7 @@ def open1c(inp_text,exp_name,gpu_numbers,pretrained_s2G_path):
|
||||
else:
|
||||
yield process_info(process_name_1c, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1c():
|
||||
global ps1c
|
||||
if (ps1c != []):
|
||||
@ -696,8 +785,11 @@ def close1c():
|
||||
ps1c = []
|
||||
return process_info(process_name_1c, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
ps1abc = []
|
||||
process_name_1abc = i18n("训练集格式化一键三连")
|
||||
|
||||
|
||||
def open1abc(inp_text, inp_wav_dir, exp_name, gpu_numbers1a, gpu_numbers1Ba, gpu_numbers1c, bert_pretrained_dir, ssl_pretrained_dir, pretrained_s2G_path):
|
||||
global ps1abc
|
||||
inp_text = my_utils.clean_path(inp_text)
|
||||
@ -707,7 +799,7 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
if (ps1abc == []):
|
||||
opt_dir = "%s/%s" % (exp_root, exp_name)
|
||||
try:
|
||||
#############################1a
|
||||
# 1a
|
||||
path_text = "%s/2-name2text.txt" % opt_dir
|
||||
if (os.path.exists(path_text) == False or (os.path.exists(path_text) == True and len(open(path_text, "r", encoding="utf8").read().strip("\n").split("\n")) < 2)):
|
||||
config = {
|
||||
@ -734,7 +826,8 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
p = Popen(cmd, shell=True)
|
||||
ps1abc.append(p)
|
||||
yield i18n("进度") + ": 1A-Doing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
for p in ps1abc:p.wait()
|
||||
for p in ps1abc:
|
||||
p.wait()
|
||||
|
||||
opt = []
|
||||
for i_part in range(all_parts): # txt_path="%s/2-name2text-%s.txt"%(opt_dir,i_part)
|
||||
@ -747,7 +840,7 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
assert len("".join(opt)) > 0, process_info(process_name_1a, "failed")
|
||||
yield i18n("进度") + ": 1A-Done", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
ps1abc = []
|
||||
#############################1b
|
||||
# 1b
|
||||
config = {
|
||||
"inp_text": inp_text,
|
||||
"inp_wav_dir": inp_wav_dir,
|
||||
@ -771,10 +864,11 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
p = Popen(cmd, shell=True)
|
||||
ps1abc.append(p)
|
||||
yield i18n("进度") + ": 1A-Done, 1B-Doing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
for p in ps1abc:p.wait()
|
||||
for p in ps1abc:
|
||||
p.wait()
|
||||
yield i18n("进度") + ": 1A-Done, 1B-Done", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
ps1abc = []
|
||||
#############################1c
|
||||
# 1c
|
||||
path_semantic = "%s/6-name2semantic.tsv" % opt_dir
|
||||
if (os.path.exists(path_semantic) == False or (os.path.exists(path_semantic) == True and os.path.getsize(path_semantic) < 31)):
|
||||
config = {
|
||||
@ -800,7 +894,8 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
p = Popen(cmd, shell=True)
|
||||
ps1abc.append(p)
|
||||
yield i18n("进度") + ": 1A-Done, 1B-Done, 1C-Doing", {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
for p in ps1abc:p.wait()
|
||||
for p in ps1abc:
|
||||
p.wait()
|
||||
|
||||
opt = ["item_name\tsemantic_audio"]
|
||||
for i_part in range(all_parts):
|
||||
@ -820,6 +915,7 @@ def open1abc(inp_text,inp_wav_dir,exp_name,gpu_numbers1a,gpu_numbers1Ba,gpu_numb
|
||||
else:
|
||||
yield process_info(process_name_1abc, "occupy"), {"__type__": "update", "visible": False}, {"__type__": "update", "visible": True}
|
||||
|
||||
|
||||
def close1abc():
|
||||
global ps1abc
|
||||
if (ps1abc != []):
|
||||
@ -831,11 +927,13 @@ def close1abc():
|
||||
ps1abc = []
|
||||
return process_info(process_name_1abc, "closed"), {"__type__": "update", "visible": True}, {"__type__": "update", "visible": False}
|
||||
|
||||
|
||||
def switch_version(version_):
|
||||
os.environ["version"] = version_
|
||||
global version
|
||||
version = version_
|
||||
if pretrained_sovits_name[int(version[-1])-1] !='' and pretrained_gpt_name[int(version[-1])-1] !='':...
|
||||
if pretrained_sovits_name[int(version[-1]) - 1] != '' and pretrained_gpt_name[int(version[-1]) - 1] != '':
|
||||
...
|
||||
else:
|
||||
gr.Warning(i18n('未下载模型') + ": " + version.upper())
|
||||
set_default()
|
||||
@ -852,23 +950,25 @@ def switch_version(version_):
|
||||
{'__type__': 'update', "interactive": False if version == "v3" else True, "value": False}, \
|
||||
{'__type__': 'update', "visible": True if version == "v3" else False}
|
||||
|
||||
if os.path.exists('GPT_SoVITS/text/G2PWModel'):...
|
||||
|
||||
if os.path.exists('GPT_SoVITS/text/G2PWModel'):
|
||||
...
|
||||
else:
|
||||
cmd = '"%s" GPT_SoVITS/download.py' % python_exec
|
||||
p = Popen(cmd, shell=True)
|
||||
p.wait()
|
||||
|
||||
|
||||
def sync(text):
|
||||
return {'__type__': 'update', 'value': text}
|
||||
|
||||
|
||||
with gr.Blocks(title="GPT-SoVITS WebUI") as app:
|
||||
gr.Markdown(
|
||||
value=
|
||||
i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.") + "<br>" + i18n("如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录LICENSE.")
|
||||
value=i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.") + "<br>" + i18n("如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录LICENSE.")
|
||||
)
|
||||
gr.Markdown(
|
||||
value=
|
||||
i18n("中文教程文档") + ": " + "https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e"
|
||||
value=i18n("中文教程文档") + ": " + "https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e"
|
||||
)
|
||||
|
||||
with gr.Tabs():
|
||||
@ -931,8 +1031,10 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
|
||||
|
||||
def change_lang_choices(key): # 根据选择的模型修改可选的语言
|
||||
return {"__type__": "update", "choices": asr_dict[key]['lang'], "value": asr_dict[key]['lang'][0]}
|
||||
|
||||
def change_size_choices(key): # 根据选择的模型修改可选的模型尺寸
|
||||
return {"__type__": "update", "choices": asr_dict[key]['size'], "value": asr_dict[key]['size'][-1]}
|
||||
|
||||
def change_precision_choices(key): # 根据选择的模型修改可选的语言
|
||||
if key == "Faster Whisper (多语种)":
|
||||
if default_batch_size <= 4:
|
||||
@ -1121,7 +1223,8 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
|
||||
|
||||
version_checkbox.change(switch_version, [version_checkbox], [pretrained_s2G, pretrained_s2D, pretrained_s1, GPT_dropdown, SoVITS_dropdown, batch_size, total_epoch, save_every_epoch, text_low_lr_rate, if_grad_ckpt, batched_infer_enabled, lora_rank])
|
||||
|
||||
with gr.TabItem(i18n("2-GPT-SoVITS-变声")):gr.Markdown(value=i18n("施工中，请静候佳音"))
|
||||
with gr.TabItem(i18n("2-GPT-SoVITS-变声")):
|
||||
gr.Markdown(value=i18n("施工中，请静候佳音"))
|
||||
|
||||
app.queue().launch( # concurrency_count=511, max_size=1022
|
||||
server_name="0.0.0.0",
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user