alumae / kaldi-gstreamer-server

Real-time full-duplex speech recognition server, based on the Kaldi toolkit and the GStreamer framwork.
BSD 2-Clause "Simplified" License
1.07k stars 342 forks source link

gstreamer is not very stable #209

Open RaniRaven opened 4 years ago

RaniRaven commented 4 years ago

There are several issues there : [1] It seems like something is getting stuck there , when using longer audios than the super-short audio given in the example. [2] The whole architecture doesn't seem like multithreaded, and therefore anything more than very simple audio sending will cause a crush and stuck of all the processes. [3] Trying to use nnet2/nnet3 models will cause a problem, as the code of kaldi-asr has added 7th parameter including OnlineCMVN state. It is not dealt. [4] Something is wrong with the way the models are implemented there, as most of the output resulting text is

alumae commented 4 years ago

[3] should be fixed now, if you recompile gst-kaldi-nnet2-online.