Hi Martin,
You're right, wow, the bug has been around for quite a long time!
It seems that the encoder part is buggy too, so tried to update your patch for that. And in the middle of it, realizing that bit shifting operations seem to complicate things (were meant optimization perhaps, but now we are about to add absolute and more checks anyway), so tried to replace them with multiplication/division operations. Also, instead of immediately stop the shifting, perhaps reducing it step by step will be better (more adaptive perhaps, as you mentioned about setting pcm shift to 1). Attached is the updated patch, please check and let us know if you have any feedback.
Thank you for the report, the analysis, and the patch.
BR,
nanang
On Tue, Sep 3, 2019 at 8:18 PM Martin Navne <martin.navne@xxxxxxxxxx> wrote:
Hi!_______________________________________________
We've been using PJMEDIA as part of a media server for voip calls. When
support for G722 was added between endpoints we noticed that the sound
level was much lower compared to G711a.
After some investigation we found that the clip detector in the G722
decoder kicked in immediately when the decoding started, and that it
seemed to be due to negative signal levels being fed into the clip
detector. The current implementation of the g722_dec_decoder return
values in the range [-16384, 16383]. The way the current clip detectoris constructed it will always detect clipping on negative values because ofthe bit representation of negative values.
With a small modification in how the signal is compared with the PCM
clip mask we avoid clip detection by negative values. The result is thatthe sound level of the decoded G722 signal is on par with G711a.The attached patch (git format-patch) show the fix we use.P.S.A PCM shift of 2 has sometimes in our environment resulted in distortion, probably due to clipping, and because of this we use a PCM shift of 1.BR,Martin NavneTelavox
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@xxxxxxxxxxxxxxx
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Attachment:
pcm_shift_g722.diff
Description: Binary data
_______________________________________________ Visit our blog: http://blog.pjsip.org pjsip mailing list pjsip@xxxxxxxxxxxxxxx http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
