mirror of
https://github.com/RVC-Boss/GPT-SoVITS.git
synced 2026-07-22 18:30:05 +08:00
7be8a59e52
为 TTS More 工作台增加远程获取训练数据的能力，并解除参考音频时长的硬编码阻断。
API (api_v2.py, 端口 9880):
- GET /models 列出训练角色(exp_name)及匹配权重
- GET /models/{name}/samples 列出训练样本(音频+参考文本), 含目录穿越校验
- GET /status 当前权重/版本/设备状态
- POST /upload_ref 上传参考音频, 文件名白名单清洗
Gradio WebUI (inference_webui.py):
- 新增「训练角色选择」区域: 模型名下拉/刷新/自动匹配权重/样本下拉/预览/应用样本
- on_auto_select_weights 采用稳健写法: 仅返回 Dropdown 值, 由已有 .change
绑定触发切换 (change_sovits_weights 是生成器, 手动消费会丢失组件更新)
时长硬编码解除 (3~10s 不再阻断):
- TTS_infer_pack/TTS.py:815 raise OSError -> print [Warning]
- inference_webui.py:853 raise OSError -> gr.Warning (建议性文案)
- inference_webui.py:1367 标签改为「推荐3~10秒」
依赖: requirements.txt 追加 python-multipart (/upload_ref 必需)
向后兼容: 现有 /tts /set_gpt_weights /set_sovits_weights 签名与返回不变,
TTS 推理 pipeline 本体逻辑未改。所有改动 py_compile 通过。
47 lines
764 B
Plaintext
47 lines
764 B
Plaintext
--no-binary=opencc
|
|
numpy<2.0
|
|
scipy
|
|
tensorboard
|
|
librosa==0.10.2
|
|
numba
|
|
pytorch-lightning>=2.4
|
|
gradio<5
|
|
ffmpeg-python
|
|
onnxruntime; platform_machine == "aarch64" or platform_machine == "arm64"
|
|
onnxruntime-gpu; platform_machine == "x86_64" or platform_machine == "AMD64"
|
|
tqdm
|
|
funasr>=1.3.7
|
|
cn2an
|
|
pypinyin
|
|
pyopenjtalk>=0.4.1
|
|
g2p_en
|
|
torchaudio
|
|
modelscope
|
|
sentencepiece
|
|
transformers>=4.43,<=4.50
|
|
peft<0.18.0
|
|
chardet
|
|
PyYAML
|
|
psutil
|
|
jieba_fast
|
|
jieba
|
|
split-lang
|
|
fast_langdetect>=0.3.1
|
|
wordsegment
|
|
rotary_embedding_torch
|
|
ToJyutping
|
|
g2pk2
|
|
ko_pron
|
|
opencc
|
|
python_mecab_ko; sys_platform != 'win32'
|
|
fastapi[standard]>=0.115.2
|
|
x_transformers
|
|
torchmetrics<=1.5
|
|
pydantic<=2.10.6
|
|
ctranslate2>=4.0,<5
|
|
av>=11
|
|
tqdm
|
|
|
|
# FastAPI file upload support for /upload_ref endpoint
|
|
python-multipart
|