Compare commits

...

5 Commits

Author SHA1 Message Date
Spr_Aachen
784d7d1cbb
Merge 35fac290ba937d032f9b635a14abf667037a1dcb into c767f0b83b998e996a4d230d86da575a03f54a3f 2026-01-05 09:48:03 +08:00
ChasonJiang
c767f0b83b
修复bug (#2704)
* 修复bug

* fallbak and bug fix
2025-12-30 16:00:21 +08:00
ChasonJiang
9080a967d5
修复采样错误 (#2703) 2025-12-30 15:21:03 +08:00
sushistack
51df9f7384
Fix model file name in README instructions (#2700) 2025-12-25 16:44:21 +08:00
Spr_Aachen
35fac290ba Improve G2PWPinyin model handling
- Add 'bert_pretrained_dir' environment variable as the highest priority for model_source
- Replace redundant g2pw initialization in chinese2.py with import from download.py
- Remove unused 'parent_directory' variable
2025-11-10 15:00:00 +08:00
4 changed files with 5 additions and 13 deletions

View File

@ -707,12 +707,11 @@ class Text2SemanticDecoder(nn.Module):
if idx == 0:
attn_mask = F.pad(attn_mask[:, :, -1].unsqueeze(-2), (0, 1), value=False)
logits = logits[:, :-1]
else:
attn_mask = F.pad(attn_mask, (0, 1), value=False)
if idx < 11: ###至少预测出10个token不然不给停止0.4s
logits = logits[:, :-1]
logits = logits[:, :-1]
samples = sample(
logits, y, top_k=top_k, top_p=top_p, repetition_penalty=repetition_penalty, temperature=temperature

View File

@ -7,7 +7,7 @@ from text.g2pw import G2PWPinyin
g2pw = G2PWPinyin(
model_dir="GPT_SoVITS/text/G2PWModel",
model_source="GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large",
model_source=os.environ.get("bert_pretrained_dir") or os.environ.get("bert_path") or "GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large",
v_to_u=False,
neutral_tone_with_five=True,
)

View File

@ -28,15 +28,8 @@ import jieba_fast.posseg as psg
is_g2pw = True # True if is_g2pw_str.lower() == 'true' else False
if is_g2pw:
# print("当前使用g2pw进行拼音推理")
from text.g2pw import G2PWPinyin, correct_pronunciation
parent_directory = os.path.dirname(current_file_path)
g2pw = G2PWPinyin(
model_dir="GPT_SoVITS/text/G2PWModel",
model_source=os.environ.get("bert_path", "GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large"),
v_to_u=False,
neutral_tone_with_five=True,
)
from text.g2pw import correct_pronunciation
from download import g2pw
rep_map = {
"": ",",

View File

@ -347,7 +347,7 @@ Use v4 from v1/v2/v3 environment:
2. Clone the latest codes from github.
3. Download v4 pretrained models (gsv-v4-pretrained/s2v4.ckpt, and gsv-v4-pretrained/vocoder.pth) from [huggingface](https://huggingface.co/lj1995/GPT-SoVITS/tree/main) and put them into `GPT_SoVITS/pretrained_models`.
3. Download v4 pretrained models (gsv-v4-pretrained/s2v4.pth, and gsv-v4-pretrained/vocoder.pth) from [huggingface](https://huggingface.co/lj1995/GPT-SoVITS/tree/main) and put them into `GPT_SoVITS/pretrained_models`.
## V2Pro Release Notes