Am Samstag, 7. Februar 2004 12:45 schrieb Damien Sandras: > Le sam 07/02/2004 ŕ 00:12, Conrad Beckert a écrit : > > Pardon for jumping in ;-) > > > > >>Something that mutes the local speakers if the level > > > > on the mic is above a given level with respect to > > what's received from peer and mutes the mic > > otherwise?<< > > It should be possible to cancel the echos: As the PC > > "knows" what it sent to the speakers it should be able > > to subtract exactly that from the stream it recieves > > from the microphone. (delayed a couple of microseconds > > - the time the sound travels from the speakers to the > > microphone + signal delay in the sound card) > > > > Doesn't Gnomemeeting, or OpenH323 rather, already have > > such a feature? I have no problems with feedback > > interference. > > It is foreseen after 1.00 to have echo cancellation. The problem with > that kind of algorithm is that it consumes a lot of CPU if you want high > quality. But we have plans about that, yes. I have some code laying on my harddisk, it is doing a 256 tap echo cancellation with double talk detection, the result are quite good. The code is not optimized nor really clean, but it does the calculation (8kHz, 16bit) on a 700MHz Duron at approx. 10% cpu usage. At the moment, the code is not ready for real use, the test program is only working on prerecorded raw audio, and I have only run minimal tests. I have uploaded some waveform plots to http://www.kawo1.rwth-aachen.de/~lurchi/echo/index.html . Problems: Code cleanup needed How to incorporate into pwlib More testing needed Do some postprocessing? Patent infringements? (Don't think so, but not sure) and last: I am busy at the moment, till 26th of Feb. ... Greetings, Stefan -- Stefan Brüns / Kastanienweg 6 - Zimmer 1206 / 52074 Aachen mailto:lurch gmx li http://www.kawo1.rwth-aachen.de/~lurchi/ phone: +49 241 169-4206 mobile: +49 160 7532733
Attachment:
pgpBGZxbA8DJi.pgp
Description: signature