Hello all, I'm just new to this list. I'm user of a voIP application which uses Speex Acoustic Echo Canceller (so only user of speex). The only parameter it has to tune is the length of the ec tail. I'm testing the echo canceller on an arm9 platform. Tests made only allow me to raise the length of the tail until 250 msec (upper the cpu power consumption is 100% and there are audio losses) getting bad results (still echo !). I have get the beta source of speex and checked that it can be compiled to use fixed point arithmetic plus assembler code for my architecture. Now I'm trying to recompile the speex library which these options and get better results. However I have some doubts: - Which is the value of the tail length to get good results? - May I specify some other options (at configuration time or may be at source) to get the speex AEC lighter, so that I can raise the tail length and get best results? - May be speex AEC tunned with another parameters to improve echo cancellation results? - AEC uses floating point in some cases (altough fixed-point specify)? (soft floating point is slow...) Sorry if this seems to be very obvious... Best regards, Victor Librado Sancho Departamento I+D ------------------------------------------------------------------------
> - Which is the value of the tail length to get good results?It depends on your setup, but if you do everything carefully, anything from 40ms to 200ms (convert to samples) can work well.> - May I specify some other options (at configuration time or may be at > source) to get the speex AEC lighter, so that I can raise the tail > length and get best results?The tail length is the main thing that affects the complexity. BTW, longer tail does not necessarily mean better results (the longer the tail, the slower the adaptation).> - May be speex AEC tunned with another parameters to improve echo > cancellation results?If the results aren't good, the most common cause is bad synchronisation between the capture and playback signals.> - AEC uses floating point in some cases (altough fixed-point specify)? > (soft floating point is slow...)No, 1.2beta2 is entirely fixed-point. If you see otherwise, you screwed up. Jean-Marc> Sorry if this seems to be very obvious... > > Best regards, > > > > Victor Librado Sancho > > Departamento I+D > ------------------------------------------------------------------------ > _______________________________________________ > Speex-dev mailing list > Speex-dev@xiph.org > http://lists.xiph.org/mailman/listinfo/speex-dev > >
Guilherme Balena Versiani
2007-Jun-26 15:05 UTC
[Speex-dev] speex AEC lighter configuration
Regarding that Speex EC, is there a way to let this parameter dynamic? I think this parameter would be different between machines, as different machines have different setups, so it would be a good idea to compute the 'echo delay' dynamically by some way during EC process. Regards, Guilherme Balena Versiani. Jean-Marc Valin wrote:>> - Which is the value of the tail length to get good results? >> > > It depends on your setup, but if you do everything carefully, anything > from 40ms to 200ms (convert to samples) can work well. > > >> - May I specify some other options (at configuration time or may be at >> source) to get the speex AEC lighter, so that I can raise the tail >> length and get best results? >> > > The tail length is the main thing that affects the complexity. BTW, > longer tail does not necessarily mean better results (the longer the > tail, the slower the adaptation). > > >> - May be speex AEC tunned with another parameters to improve echo >> cancellation results? >> > > If the results aren't good, the most common cause is bad synchronisation > between the capture and playback signals. > > >> - AEC uses floating point in some cases (altough fixed-point specify)? >> (soft floating point is slow...) >> > > No, 1.2beta2 is entirely fixed-point. If you see otherwise, you screwed up. > > Jean-Marc > > >> Sorry if this seems to be very obvious... >> >> Best regards, >> >> >> >> Victor Librado Sancho >> >> Departamento I+D >> ------------------------------------------------------------------------ >> _______________________________________________ >> Speex-dev mailing list >> Speex-dev@xiph.org >> http://lists.xiph.org/mailman/listinfo/speex-dev >> >> >> > _______________________________________________ > Speex-dev mailing list > Speex-dev@xiph.org > http://lists.xiph.org/mailman/listinfo/speex-dev