[allow regression][Non-BE][Split Non-BE] Add FIX_2000_NON_LINEARITY_OVERSHOOT to address SWB TBE overshoot in issue 2000
requested to merge 2000-bwe-artifact-for-omasa-1-object-16-lkfs-26-lkfs-24-4-and-48-kbps-c12-and-c14-in-p800-18 into main
Closes #2000 (closed)
I found that issue 2000 was triggered because of the mismatch between the float and the BASOP in NONBE_1328_FIX_NON_LINEARITY. This brings the BASOP code in line with the float code, and it seems to help for the cases observed here.
Note that I changed the sc_factor to be a shift parameter instead of a factor. This way we can do a shift instead of a multiplication when checking the condition. I hope I got it right, but at least it fixed the problem on my sample.
I am not sure who should review. The code is from QC, but NONBE_1328_FIX_NON_LINEARITY is from VoiceAge. I think also FhG colleagues looked at this. Pinging @multrus, @vaillancour, @jelinek, @vasilache.