modelscope / FunASR

A Fundamental End-to-End Speech Recognition Toolkit and Open Source SOTA Pretrained Models, Supporting Speech Recognition, Voice Activity Detection, Text Post-processing etc.
https://www.funasr.com
Other
6.47k stars 688 forks source link

paraformer-online模型和funasr的在线服务测试结果不一致 #1630

Closed wwfcnu closed 5 months ago

wwfcnu commented 6 months ago

直接通过modelscope库,测试paraformer-online 模型的结果和通过funasr在线服务(mode=online)测试的结果不一样, funasr的结果是要比前者好的,这是为什么,funasr的在线服务用的模型也是paraformer-online 呀

LauraGPT commented 6 months ago

Please check the differences of config

wwfcnu commented 6 months ago

是不是funasr的在线docker服务对结果做了后处理,直接通过paraformer-online模型转录的结果中包含很多重复的字

LauraGPT commented 6 months ago

It is a bug. Please update funasr and use the AutoModel

wwfcnu commented 5 months ago

It is a bug. Please update funasr and use the AutoModel 改成fuansr的Automodel,结果就和在线服务对齐了