Voice separation on multiple inputs via microphone

[MuhammadRashid]

Hi Nickolay, Here is a scenario, suppose an audio is playing in a user's Andorid device and meanwhile an announcement occurs outside, now microphone is capturing multiple inputs. How can we recognise and separate these two sources (audio playing in a user's device and announcement outside)?

[nshmyrev]

I didn't investigate this deeply, it should be possible to record without capturing playback, but no idea how to do it properly.

[huynguyen82]

Hi Nickolay, The model directory contains HCLr.fst and Gr.fst, so can you please show me how to produce them? I found a script to produce HCL.fst and G.fst at https://github.com/opendcd/opendcd/blob/master/script/makeclevel.sh. Can i use it? Thanks!

[nshmyrev]

@huynguyen82 please create a separate issue, do not post into other unrelated issue. Graph is created with mkgraph_lookahead_arpa.sh, see https://github.com/kaldi-asr/kaldi/pull/3616