Hi We are working on a speaker phone system using PJSIP and Speex Speech processing API on an ARM platform. Currently we have spent about a month on getting the AEC to work properly and we have worked through the most common causes of problems (such as clock drift, synchronization problems and non-linearity's in echo path). Now we achieve ERLE of about 17 dB which tells me that the AEC is doing something however we are subjectively not satisfied. Some references of ERLE taken from this mailing list tells that 15 dB [1,2] is not good enough and that 30 dB would be fine [3]. My question is if 17 dB is in line with what is expected from the Speex algorithm? You can find some dumped files on this link: https://extranet.svep.se/AEC_dump_100ms.zip These files are dumped in mdf.c when running on the ARM platform with an IP-phone conversation active (only far end talking). All preprocessors have been disabled. We have been using a high quality measurement microphone and a fairly high quality PC speaker in the test setup. The result is similar when we run the signals offline through testecho.c. We have been using PJSIP revision 2709 from svn which contains Speex-1.2beta3. Best Regards /Johan Nilsson Some links: [1] http://lists.xiph.org/pipermail/speex-dev/2007-March/005390.html [2] http://lists.xiph.org/pipermail/speex-dev/2005-June/003352.html [3] http://lists.xiph.org/pipermail/speex-dev/2005-June/003358.html -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/speex-dev/attachments/20090527/389060fb/attachment.htm
Hi Jean-Marc & all, I'm currently working on a wav file with very low volume. This wav also contains a DC component. Using testdenoise with modification (AGC turned on with 20000.0 value) I'm able to increase volume, but a new irregular DC component appears again due to the AGC. (seems to increase at the same time the AGC gain is increasing) Using testdenoise again on the output, but without the AGC, the preprocessor remove the newly introduced DC component. Note that the input has really low volume which is most probably the main issue here. Still this is showing that using the AGC potentially introduce DC component? Any comment? Tks, Aymeric MOIZARD / ANTISIP amsip - http://www.antisip.com osip2 - http://www.osip.org eXosip2 - http://savannah.nongnu.org/projects/exosip/
Quoting Johan Nilsson <han at svep.se>:> Now we achieve ERLE of about 17 dB which tells me that the AEC is doing > something however we are subjectively not satisfied. Some references of ERLE > taken from this mailing list tells that 15 dB [1,2] is not good enough and > that 30 dB would be fine [3]. > > My question is if 17 dB is in line with what is expected from the Speex > algorithm?An ERLE of 17 dB *without* the preprocessor is pretty much what you can expect from the Speex AEC. Of course, it depends a lot on the conditions, recording equipment, double-talk, ... See Fig. 5 of this paper: http://people.xiph.org/~jm/papers/valin_tasl2007.pdf and Fig. 2 of this one: http://people.xiph.org/~jm/papers/valin_icassp2007.pdf for some ERLE curves with the Speex AEC. As you say, 17 dB is not enough to really remove the echo, so you probably want to apply the preprocessor on the output to remove the residual echo (see example and manual for how to link the two). This should be able to do the job. Jean-Marc
I'm getting confused, what's the problem exactly? The AGC doesn't remove the DC or it adds one? Jean-Marc Aymeric Moizard a ?crit :> Hi Jean-Marc & all, > > I'm currently working on a wav file with very low volume. > This wav also contains a DC component. > > Using testdenoise with modification (AGC turned on with > 20000.0 value) I'm able to increase volume, but a new > irregular DC component appears again due to the AGC. > (seems to increase at the same time the AGC gain is increasing) > > Using testdenoise again on the output, but without the > AGC, the preprocessor remove the newly introduced DC > component. > > Note that the input has really low volume which is most > probably the main issue here. Still this is showing that > using the AGC potentially introduce DC component? Any > comment? > > Tks, > Aymeric MOIZARD / ANTISIP > amsip - http://www.antisip.com > osip2 - http://www.osip.org > eXosip2 - http://savannah.nongnu.org/projects/exosip/ > > _______________________________________________ > Speex-dev mailing list > Speex-dev at xiph.org > http://lists.xiph.org/mailman/listinfo/speex-dev > >