Detailed documentation of CTC decoder code

I am planning to understand the CTC decoder code and was wondering if there is any detailed documentation of it. Is this implemention of any algorithm I can refer to.

Also is there any change in CTC decoder code between version 5 and 6 of deepspeech?

Thanks in advance.

git log --reverse --stat v0.5.0...v0.6.1 -- native_client/ctcdecode

A nice intro to CTC is available on distill: https://distill.pub/2017/ctc/

Also the video from Bhiksha Raj is very good to get started: https://www.youtube.com/watch?v=c86gfVGcvh4

Thanks Ruben and dwn.

1 Like