Hi, After some search, I found the problem I have pointed is not new. This problem has been addressed before in a series of threads (not well organized in the list history for some reason): http://lists.pjsip.org/pipermail/pjsip_lists.pjsip.org/2008-September/004833.html As a matter of fact, the quality degradation I see is quite similar to the one pointed by Massimiliano and the first explanation given by Benny matches exactly what I have seen in my debugs. In the beginning of the call, frames are inserted in the jitter buffer in a burst and jitter buffer reacts by dropping the excessive frames. Even though this problem have been addressed before, I does not seem to have been resolved. As I said before, in a call from A to B, MOS measured at A is lower than at B (which is 4.4 as expected). I don't think this kind of behavior should be accepted as it clearly represents a quality degradation in a great number of calls and can interfere directly on PESQ results. The problem is not exactly located in the jitter buffer, but probably in ioqueue who is causing this burst of insertions in the buffer. Since this problem only occurs in one host of peer-to-peer call, I think it is a matter of comparing the caller and the callee behavior. I will try to investigate this matter in my spare time, but would also like to listen to some more experienced opinions, of course. =D -- []'s Andr? de Abrantes On Tue, Jun 29, 2010 at 18:54, Andr? Abrantes <andreadps at gmail.com> wrote: > Hi, > > I have been studying voice quality using pjsua and came across a nice > tricky "feature" of pjmedia's jitter buffer. Either using progressive > discard or not, in the begging of the call, some frames are discarded > unnecessarily (20~25). > > It seems to happen because there is a big number of frames waiting in > the buffer right before the jitter buffer's state changes from > initializing to processing. All of this, however, does not happen all > the time. To be more specific, if pjsua A calls directly > (peer-to-peer) to pjsua B, this event can be noticed only in A. > > I have been trying to find a nice way of minimizing this effect but > have not found anything worth until now. As I am not exactly an expert > in pjsip, I decided to ask around here if this kind of discard has > been noticed before. Should I open a ticket for this? > > Thanks! > > -- > []'s > Andr? de Abrantes >
