mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2025-10-07 15:19:59 +08:00
372 lines
14 KiB
Python
372 lines
14 KiB
Python
import config
|
||
import sys,os
|
||
import gradio as gr
|
||
import torch
|
||
import numpy as np
|
||
import librosa,torch
|
||
from transformers import AutoModelForMaskedLM, AutoTokenizer
|
||
from feature_extractor import cnhubert
|
||
from module.models import SynthesizerTrn
|
||
from AR.models.t2s_lightning_module import Text2SemanticLightningModule
|
||
from text import cleaned_text_to_sequence
|
||
from text.cleaner import clean_text
|
||
from time import time as ttime
|
||
from module.mel_processing import spectrogram_torch
|
||
from my_utils import load_audio
|
||
from tools.i18n.i18n import I18nAuto
|
||
|
||
hps = None
|
||
ssl_model = None
|
||
vq_model = None
|
||
t2s_model = None
|
||
is_half = config.is_half
|
||
hz = 50
|
||
max_sec = None
|
||
top_k = None
|
||
|
||
#后期可能将这里个path分离成变量
|
||
bert_path = config.bert_path
|
||
cnhubert_base_path = config.cnhubert_path
|
||
cnhubert.cnhubert_base_path = cnhubert_base_path
|
||
device = "cuda" #不确定能否支持cpu,先预留
|
||
|
||
tokenizer = None
|
||
bert_model = None
|
||
|
||
i18n = I18nAuto()
|
||
cwd = os.getcwd()
|
||
sys.path.append(cwd)
|
||
|
||
SUPPORT_LANGUAGE = [i18n("中文"),i18n("英文"),i18n("日文")]
|
||
|
||
dict_language={
|
||
i18n("中文"):"zh",
|
||
i18n("英文"):"en",
|
||
i18n("日文"):"ja"
|
||
}
|
||
|
||
def read_model_path(model_type):
|
||
model_list = []
|
||
if model_type == config.MODEL_TYPE_GPT:
|
||
folder_path = os.path.join(cwd,config.MODEL_FOLDER_PATH_GPT)
|
||
file_type = ".ckpt"
|
||
elif model_type == config.MODEL_TYPE_SOVITS:
|
||
folder_path = os.path.join(cwd,config.MODEL_FOLDER_PATH_SOVITS)
|
||
file_type = ".pth"
|
||
for root, dirs, files in os.walk(folder_path):
|
||
for file_name in files:
|
||
if file_name.endswith(file_type):
|
||
file_path = os.path.join(root, file_name)
|
||
model_list.append((file_name,file_path))
|
||
return model_list
|
||
|
||
def refresh_model_list():
|
||
gpt_choices = read_model_path(config.MODEL_TYPE_GPT)
|
||
sovits_choices = read_model_path(config.MODEL_TYPE_SOVITS)
|
||
return gr.Dropdown(choices=sorted(gpt_choices),value=gpt_choices[0]if len(gpt_choices)>0 else "",interactive=True),gr.Dropdown(choices=sorted(sovits_choices),value=sovits_choices[0]if len(sovits_choices)>0 else None,interactive=True)
|
||
|
||
def get_bert_feature(text, word2ph):
|
||
global tokenizer,bert_model
|
||
with torch.no_grad():
|
||
inputs = tokenizer(text, return_tensors="pt")
|
||
for i in inputs:
|
||
inputs[i] = inputs[i].to(device) #####输入是long不用管精度问题，精度随bert_model
|
||
res = bert_model(**inputs, output_hidden_states=True)
|
||
res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()[1:-1]
|
||
assert len(word2ph) == len(text)
|
||
phone_level_feature = []
|
||
for i in range(len(word2ph)):
|
||
repeat_feature = res[i].repeat(word2ph[i], 1)
|
||
phone_level_feature.append(repeat_feature)
|
||
phone_level_feature = torch.cat(phone_level_feature, dim=0)
|
||
# if(is_half==True):phone_level_feature=phone_level_feature.half()
|
||
return phone_level_feature.T
|
||
|
||
class DictToAttrRecursive(dict):
|
||
def __init__(self, input_dict):
|
||
super().__init__(input_dict)
|
||
for key, value in input_dict.items():
|
||
if isinstance(value, dict):
|
||
value = DictToAttrRecursive(value)
|
||
self[key] = value
|
||
setattr(self, key, value)
|
||
|
||
def __getattr__(self, item):
|
||
try:
|
||
return self[item]
|
||
except KeyError:
|
||
raise AttributeError(f"Attribute {item} not found")
|
||
|
||
def __setattr__(self, key, value):
|
||
if isinstance(value, dict):
|
||
value = DictToAttrRecursive(value)
|
||
super(DictToAttrRecursive, self).__setitem__(key, value)
|
||
super().__setattr__(key, value)
|
||
|
||
def __delattr__(self, item):
|
||
try:
|
||
del self[item]
|
||
except KeyError:
|
||
raise AttributeError(f"Attribute {item} not found")
|
||
|
||
def get_spepc(hps, filename):
|
||
audio = load_audio(filename, int(hps.data.sampling_rate))
|
||
audio = torch.FloatTensor(audio)
|
||
audio_norm = audio
|
||
audio_norm = audio_norm.unsqueeze(0)
|
||
spec = spectrogram_torch(
|
||
audio_norm,
|
||
hps.data.filter_length,
|
||
hps.data.sampling_rate,
|
||
hps.data.hop_length,
|
||
hps.data.win_length,
|
||
center=False,
|
||
)
|
||
return spec
|
||
|
||
def load_models(sovits_path, gpt_path):
|
||
global tokenizer,bert_model,hps,ssl_model,vq_model,t2s_model,is_half,hz,max_sec,top_k
|
||
print(f"SoVITS model path: {sovits_path}")
|
||
print(f"GPT model path: {gpt_path}")
|
||
|
||
if sovits_path is None or gpt_path is None:
|
||
print("Choose both of two models before loading")
|
||
return "请正确选择两个模型",gr.Button(interactive=False)
|
||
|
||
torch.cuda.empty_cache()
|
||
|
||
tokenizer = AutoTokenizer.from_pretrained(bert_path)
|
||
bert_model = AutoModelForMaskedLM.from_pretrained(bert_path)
|
||
if is_half == True:
|
||
bert_model = bert_model.half().to(device)
|
||
else:
|
||
bert_model = bert_model.to(device)
|
||
|
||
dict_s2=torch.load(sovits_path,map_location="cpu")
|
||
hps=dict_s2["config"]
|
||
hps = DictToAttrRecursive(hps)
|
||
hps.model.semantic_frame_rate = "25hz"
|
||
|
||
dict_s1 = torch.load(gpt_path, map_location="cpu")
|
||
dict_s1_config = dict_s1["config"]
|
||
max_sec = dict_s1_config["data"]["max_sec"]
|
||
top_k=dict_s1_config["inference"]["top_k"]
|
||
|
||
ssl_model = cnhubert.get_model()
|
||
if is_half == True:
|
||
ssl_model = ssl_model.half().to(device)
|
||
else:
|
||
ssl_model = ssl_model.to(device)
|
||
|
||
vq_model = SynthesizerTrn(
|
||
hps.data.filter_length // 2 + 1,
|
||
hps.train.segment_size // hps.data.hop_length,
|
||
n_speakers=hps.data.n_speakers,
|
||
**hps.model
|
||
)
|
||
if is_half:
|
||
vq_model = vq_model.half().to(device)
|
||
else:
|
||
vq_model = vq_model.to(device)
|
||
vq_model.eval()
|
||
print(vq_model.load_state_dict(dict_s2["weight"], strict=False))
|
||
|
||
t2s_model = Text2SemanticLightningModule(dict_s1_config, "ojbk", is_train=False)
|
||
t2s_model.load_state_dict(dict_s1["weight"])
|
||
if is_half == True:
|
||
t2s_model = t2s_model.half()
|
||
t2s_model = t2s_model.to(device)
|
||
t2s_model.eval()
|
||
total = sum([param.nelement() for param in t2s_model.parameters()])
|
||
print("Number of parameter: %.2fM" % (total / 1e6))
|
||
|
||
#加载模型成功
|
||
return "模型加载成功",gr.Button(interactive=True)
|
||
|
||
|
||
def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language):
|
||
global hps,ssl_model,vq_model,t2s_model,is_half,hz,max_sec,top_k
|
||
t0 = ttime()
|
||
prompt_text = prompt_text.strip("\n")
|
||
prompt_language, text = prompt_language, text.strip("\n")
|
||
zero_wav = np.zeros(
|
||
int(hps.data.sampling_rate * 0.3),
|
||
dtype=np.float16 if is_half else np.float32,
|
||
)
|
||
with torch.no_grad():
|
||
wav16k, sr = librosa.load(ref_wav_path, sr=16000)
|
||
wav16k = torch.from_numpy(wav16k)
|
||
zero_wav_torch = torch.from_numpy(zero_wav)
|
||
if is_half == True:
|
||
wav16k = wav16k.half().to(device)
|
||
zero_wav_torch = zero_wav_torch.half().to(device)
|
||
else:
|
||
wav16k = wav16k.to(device)
|
||
zero_wav_torch = zero_wav_torch.to(device)
|
||
wav16k=torch.cat([wav16k,zero_wav_torch])
|
||
ssl_content = ssl_model.model(wav16k.unsqueeze(0))[
|
||
"last_hidden_state"
|
||
].transpose(
|
||
1, 2
|
||
) # .float()
|
||
codes = vq_model.extract_latent(ssl_content)
|
||
prompt_semantic = codes[0, 0]
|
||
t1 = ttime()
|
||
prompt_language = dict_language[prompt_language]
|
||
text_language = dict_language[text_language]
|
||
phones1, word2ph1, norm_text1 = clean_text(prompt_text, prompt_language)
|
||
phones1 = cleaned_text_to_sequence(phones1)
|
||
texts = text.split("\n")
|
||
audio_opt = []
|
||
for text in texts:
|
||
# 解决输入目标文本的空行导致报错的问题
|
||
if (len(text.strip()) == 0):
|
||
continue
|
||
phones2, word2ph2, norm_text2 = clean_text(text, text_language)
|
||
phones2 = cleaned_text_to_sequence(phones2)
|
||
if prompt_language == "zh":
|
||
bert1 = get_bert_feature(norm_text1, word2ph1).to(device)
|
||
else:
|
||
bert1 = torch.zeros(
|
||
(1024, len(phones1)),
|
||
dtype=torch.float16 if is_half == True else torch.float32,
|
||
).to(device)
|
||
if text_language == "zh":
|
||
bert2 = get_bert_feature(norm_text2, word2ph2).to(device)
|
||
else:
|
||
bert2 = torch.zeros((1024, len(phones2))).to(bert1)
|
||
bert = torch.cat([bert1, bert2], 1)
|
||
|
||
all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
|
||
bert = bert.to(device).unsqueeze(0)
|
||
all_phoneme_len = torch.tensor([all_phoneme_ids.shape[-1]]).to(device)
|
||
prompt = prompt_semantic.unsqueeze(0).to(device)
|
||
t2 = ttime()
|
||
with torch.no_grad():
|
||
# pred_semantic = t2s_model.model.infer(
|
||
pred_semantic, idx = t2s_model.model.infer_panel(
|
||
all_phoneme_ids,
|
||
all_phoneme_len,
|
||
prompt,
|
||
bert,
|
||
# prompt_phone_len=ph_offset,
|
||
top_k=top_k,
|
||
early_stop_num=hz * max_sec,
|
||
)
|
||
t3 = ttime()
|
||
# print(pred_semantic.shape,idx)
|
||
pred_semantic = pred_semantic[:, -idx:].unsqueeze(
|
||
0
|
||
) # .unsqueeze(0)#mq要多unsqueeze一次
|
||
refer = get_spepc(hps, ref_wav_path) # .to(device)
|
||
if is_half == True:
|
||
refer = refer.half().to(device)
|
||
else:
|
||
refer = refer.to(device)
|
||
# audio = vq_model.decode(pred_semantic, all_phoneme_ids, refer).detach().cpu().numpy()[0, 0]
|
||
audio = (
|
||
vq_model.decode(
|
||
pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refer
|
||
)
|
||
.detach()
|
||
.cpu()
|
||
.numpy()[0, 0]
|
||
) ###试试重建不带上prompt部分
|
||
audio_opt.append(audio)
|
||
audio_opt.append(zero_wav)
|
||
t4 = ttime()
|
||
print("%.3f\t%.3f\t%.3f\t%.3f" % (t1 - t0, t2 - t1, t3 - t2, t4 - t3))
|
||
yield hps.data.sampling_rate, (np.concatenate(audio_opt, 0) * 32768).astype(
|
||
np.int16
|
||
)
|
||
|
||
splits = {
|
||
"，",
|
||
"。",
|
||
"？",
|
||
"！",
|
||
",",
|
||
".",
|
||
"?",
|
||
"!",
|
||
"~",
|
||
":",
|
||
"：",
|
||
"—",
|
||
"…",
|
||
} # 不考虑省略号
|
||
|
||
|
||
def split(todo_text):
|
||
todo_text = todo_text.replace("……", "。").replace("——", "，")
|
||
if todo_text[-1] not in splits:
|
||
todo_text += "。"
|
||
i_split_head = i_split_tail = 0
|
||
len_text = len(todo_text)
|
||
todo_texts = []
|
||
while 1:
|
||
if i_split_head >= len_text:
|
||
break # 结尾一定有标点，所以直接跳出即可，最后一段在上次已加入
|
||
if todo_text[i_split_head] in splits:
|
||
i_split_head += 1
|
||
todo_texts.append(todo_text[i_split_tail:i_split_head])
|
||
i_split_tail = i_split_head
|
||
else:
|
||
i_split_head += 1
|
||
return todo_texts
|
||
|
||
|
||
def start_webui():
|
||
|
||
ngpu = torch.cuda.device_count()
|
||
gpu_list = []
|
||
for i in range(ngpu):
|
||
gpu_list.append((torch.cuda.get_device_name(i),i))
|
||
print(gpu_list)
|
||
|
||
gpt_choices = read_model_path(config.MODEL_TYPE_GPT)
|
||
sovits_choices = read_model_path(config.MODEL_TYPE_SOVITS)
|
||
|
||
with gr.Blocks() as demo:
|
||
with gr.Row():
|
||
message_text = gr.Textbox("信息",interactive=False)
|
||
with gr.Accordion(label="设备"):
|
||
with gr.Row():
|
||
cuda_device_index = gr.Dropdown(choices=gpu_list,value=0 if len(gpu_list)>0 else None,label="CUDA设备",interactive=True)
|
||
with gr.Accordion(label="模型"):
|
||
with gr.Row():
|
||
gpt_dropdown = gr.Dropdown(choices=sorted(gpt_choices),value=gpt_choices[0][1]if len(gpt_choices)>0 else None,label="选择GPT模型",interactive=True)
|
||
sovits_dropdown = gr.Dropdown(choices=sorted(sovits_choices),value=sovits_choices[0][1]if len(sovits_choices)>0 else None,label="选择SoVITS模型",interactive=True)
|
||
with gr.Row():
|
||
model_load_button = gr.Button("加载模型",variant="primary")
|
||
model_refresh_button = gr.Button("刷新模型", variant="secondary")
|
||
with gr.Accordion(label="参考"):
|
||
with gr.Group():
|
||
with gr.Row():
|
||
with gr.Row():
|
||
ref_wav_path = gr.Audio(label="参考音频", type="filepath", scale=3)
|
||
ref_language = gr.Dropdown(choices=SUPPORT_LANGUAGE,value=i18n("中文"),label="参考语种",interactive=True,min_width=50, scale=1)
|
||
ref_text = gr.TextArea(label="参考文本",scale=1)
|
||
with gr.Row():
|
||
output_language = gr.Dropdown(choices=SUPPORT_LANGUAGE,value=i18n("中文"),label="合成语种",interactive=True, scale=2)
|
||
preprocess_output_text_button = gr.Button("合成文本预处理",variant="primary",scale=3)
|
||
inference_button = gr.Button(i18n("合成语音"), interactive=False,variant="primary")
|
||
output_text = gr.TextArea(label="合成文本",interactive=True)
|
||
output_audio = gr.Audio(label="输出结果")
|
||
model_load_button.click(load_models,[gpt_dropdown,sovits_dropdown],[message_text,inference_button])
|
||
model_refresh_button.click(refresh_model_list,[],[gpt_dropdown,sovits_dropdown])
|
||
inference_button.click(
|
||
get_tts_wav,
|
||
[ref_wav_path, ref_text, ref_language, output_text, output_language],
|
||
[output_audio],
|
||
)
|
||
demo.queue(max_size=1022).launch(server_port=2777)
|
||
|
||
if __name__ == "__main__":
|
||
start_webui()
|
||
|
||
|
||
|
||
|
||
|