Shanyong Yu

Hi: Because I modified the kaldi source code, I want to install the pykaldi-cpu from source.What should i do,? thanks!

Dear All： I encountered a problem, I don’t know if it can be avoided. E.g: `NnetLatticeFasterOnlineRecognizer.decode()`, I use it to decode a wav file, and when I call it once,...

NPU Error

5

Conformer模型使用Paraformer的tokenizer，会导致grad_norm NaN，进而NPU会报Inner Error，但使用GPU训练一切正常  ### 复现方式 - 拉最新代码 - 修改wenet/examples/aishell/s0/run_npu.sh 下面 conf/train_conformer.yaml，将tokenizer改成paraformer ``` tokenizer: paraformer tokenizer_conf: seg_dict_path: exp/paraformer/large/seg_dict special_tokens: : 0 : 2 : 1 : 8403 symbol_table_path: exp/paraformer/large/units.txt ```...

API参数没注释

4

```py ChatTTS.Chat.infer( self, text, stream=False, lang=None, skip_refine_text=False, refine_text_only=False, use_decoder=True, do_text_normalization=True, do_homophone_replacement=True, params_refine_text=, params_infer_code=, ) ``` 其中如下几个参数，不理解是什么意思，具体是什么作用 ```py lang=None, skip_refine_text=False, refine_text_only=False, use_decoder=True, ```