From 3ced402f7459954ebc10dbd981caa6a4a96ae009 Mon Sep 17 00:00:00 2001 From: Intro Date: Thu, 13 Mar 2025 05:02:46 +0800 Subject: [PATCH 1/4] =?UTF-8?q?=E6=B7=BB=E5=8A=A0=E8=87=AA=E5=AE=9A?= =?UTF-8?q?=E4=B9=89=E8=AF=AD=E8=A8=80=E8=AF=86=E5=88=AB=E5=99=A8=E5=BC=80?= =?UTF-8?q?=E5=85=B3?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- config.py | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/config.py b/config.py index 1f74128..1a90cd5 100644 --- a/config.py +++ b/config.py @@ -15,6 +15,9 @@ bert_path = "GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large" pretrained_sovits_path = "GPT_SoVITS/pretrained_models/s2G488k.pth" pretrained_gpt_path = "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt" +# 部分模型的启用 +enable_costum_langdetect = False + exp_root = "logs" python_exec = sys.executable or "python" if torch.cuda.is_available(): @@ -54,6 +57,8 @@ class Config: self.pretrained_sovits_path = pretrained_sovits_path self.pretrained_gpt_path = pretrained_gpt_path + self.enable_costum_langdetect = enable_costum_langdetect + self.exp_root = exp_root self.python_exec = python_exec self.infer_device = infer_device From 86f7a7d98515fffe1bf90fa7f6fe70fc27134aa4 Mon Sep 17 00:00:00 2001 From: Intro Date: Thu, 13 Mar 2025 05:04:23 +0800 Subject: [PATCH 2/4] =?UTF-8?q?=E6=B7=BB=E5=8A=A0=E5=B7=A5=E4=BD=9C?= =?UTF-8?q?=E7=9B=AE=E5=BD=95=E5=AF=BC=E5=85=A5=EF=BC=8C=E7=A1=AE=E4=BF=9D?= =?UTF-8?q?=E6=A8=A1=E5=9D=97=E5=AF=BC=E5=85=A5=E6=AD=A3=E5=B8=B8?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- GPT_SoVITS/BigVGAN/bigvgan.py | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/GPT_SoVITS/BigVGAN/bigvgan.py b/GPT_SoVITS/BigVGAN/bigvgan.py index 214672e..6f52d8f 100644 --- a/GPT_SoVITS/BigVGAN/bigvgan.py +++ b/GPT_SoVITS/BigVGAN/bigvgan.py @@ -14,6 +14,12 @@ import torch.nn as nn from torch.nn import Conv1d, ConvTranspose1d from torch.nn.utils import weight_norm, remove_weight_norm +import sys + +# 确保工作目录包含该目录，否则无法导入 activations +now_dir = os.getcwd() +sys.path.append("%s/GPT_SoVITS/BigVGAN" % (now_dir)) + import activations from utils0 import init_weights, get_padding from alias_free_activation.torch.act import Activation1d as TorchActivation1d @@ -431,7 +437,8 @@ class BigVGAN( """Load Pytorch pretrained weights and return the loaded model.""" # Download and load hyperparameters (h) used by BigVGAN - if os.path.isdir(model_id): + isLocal = os.path.isdir(model_id) + if isLocal: # print("Loading config.json from local directory") config_file = os.path.join(model_id, "config.json") else: From 46d993ed04e56f11d924280468065a4f1d65be41 Mon Sep 17 00:00:00 2001 From: Intro Date: Thu, 13 Mar 2025 05:07:06 +0800 Subject: [PATCH 3/4] =?UTF-8?q?=E6=9B=B4=E6=94=B9fast=5Flangdetect?= =?UTF-8?q?=E4=B8=BA=E5=8F=AF=E9=85=8D=E7=BD=AE=E9=A1=B9=EF=BC=8C=E5=8F=AF?= =?UTF-8?q?=E5=9C=A8config.py=E4=B8=AD=E8=AE=BE=E7=BD=AE=E6=98=AF=E5=90=A6?= =?UTF-8?q?=E4=BD=BF=E7=94=A8=E8=87=AA=E5=AE=9A=E4=B9=89=E8=AF=86=E5=88=AB?= =?UTF-8?q?=E5=99=A8=20[=E5=90=90=E6=A7=BD=E4=B8=80=E4=B8=8B=E6=80=8E?= =?UTF-8?q?=E4=B9=88=E5=81=9A=E6=B5=8B=E8=AF=95=E7=9A=84=E6=96=87=E6=9C=AC?= =?UTF-8?q?=E6=98=AFgo=E6=89=B9=E5=95=8A(=E5=96=9C)]?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- GPT_SoVITS/text/LangSegmenter/langsegmenter.py | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/GPT_SoVITS/text/LangSegmenter/langsegmenter.py b/GPT_SoVITS/text/LangSegmenter/langsegmenter.py index c558348..c82de2e 100644 --- a/GPT_SoVITS/text/LangSegmenter/langsegmenter.py +++ b/GPT_SoVITS/text/LangSegmenter/langsegmenter.py @@ -1,14 +1,22 @@ import logging import re +from pathlib import Path + # jieba静音 import jieba jieba.setLogLevel(logging.CRITICAL) -# 更改fast_langdetect大模型位置 -from pathlib import Path +# 加载配置 +import sys +sys.path.append("..") +from config import Config +config = Config() + import fast_langdetect -fast_langdetect.infer._default_detector = fast_langdetect.infer.LangDetector(fast_langdetect.infer.LangDetectConfig(cache_dir=Path(__file__).parent.parent.parent / "pretrained_models" / "fast_langdetect")) +# 更改fast_langdetect大模型位置 +if config.enable_costum_langdetect: + fast_langdetect.infer._default_detector = fast_langdetect.infer.LangDetector(fast_langdetect.infer.LangDetectConfig(cache_dir=Path(__file__).parent.parent.parent / "pretrained_models" / "fast_langdetect")) from split_lang import LangSplitter From 610825769a998fe2f17c3ecce4fd31ae02604299 Mon Sep 17 00:00:00 2001 From: Intro Date: Thu, 13 Mar 2025 05:24:39 +0800 Subject: [PATCH 4/4] =?UTF-8?q?=E8=87=AA=E5=8A=A8=E4=B8=8B=E8=BD=BDnvidia/?= =?UTF-8?q?bigvgan=5Fv2=5F24khz=5F100band=5F256x?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- api.py | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/api.py b/api.py index d92d9c8..d543e73 100644 --- a/api.py +++ b/api.py @@ -172,6 +172,8 @@ from tools.my_utils import load_audio import config as global_config import logging import subprocess +from pathlib import Path +import tqdm class DefaultRefer: @@ -199,6 +201,12 @@ def is_full(*items): # 任意一项为空返回False def init_bigvgan(): + # 判断是否存在bigvgan模型 `nvidia/bigvgan_v2_24khz_100band_256x` 不存在则下载 + if not (Path(__file__).parent / "GPT_SoVITS/pretrained_models/models--nvidia--bigvgan_v2_24khz_100band_256x").exists(): + print("Downloading bigvgan model... from [nvidia/bigvgan_v2_24khz_100band_256x]") + from huggingface_hub import snapshot_download + snapshot_download(repo_id="nvidia/bigvgan_v2_24khz_100band_256x", repo_type="model", revision="main", cache_dir="GPT_SoVITS/pretrained_models") + global bigvgan_model from BigVGAN import bigvgan bigvgan_model = bigvgan.BigVGAN.from_pretrained("%s/GPT_SoVITS/pretrained_models/models--nvidia--bigvgan_v2_24khz_100band_256x" % (now_dir,), use_cuda_kernel=False) # if True, RuntimeError: Ninja is required to load C++ extensions @@ -363,7 +371,7 @@ def get_gpt_weights(gpt_path): gpt = Gpt(max_sec, t2s_model) return gpt -def change_gpt_sovits_weights(gpt_path,sovits_path): +def change_gpt_sovits_weights(gpt_path, sovits_path): try: gpt = get_gpt_weights(gpt_path) sovits = get_sovits_weights(sovits_path)