Hi Klaus, If I get your question right this time the point is in the perceptual model utilized in a voice quality assessment software. Basically, the best way is to use both: MOS generated by our software or P.862/P.563 together with typical VoIP characteristics. Our software like ITU standards (if I may compare them here) implement a perceptual voice quality assessment model that produces MOS scores according to how a human percepts the audio thus if packets were lost or "recovered" that will lead to MOS score decrease or increase. Hope I answered to your comment. Thanks! ----- Original Message ----- From: "Klaus Darilion" <klaus.mailinglists@xxxxxxxxx> To: "pjsip list" <pjsip at lists.pjsip.org> Cc: "Sevana Oy" <sales at sevana.fi> Sent: Wednesday, May 12, 2010 11:50 AM Subject: Re: Turn your free SIP softphone into a voice quality monitoring instrument with Sevana?s NIQA application > Hi! > > Am 12.05.2010 06:37, schrieb Sevana Oy: >>> Does somebody know how pjsip writes the wavefile? Will it be written >>> exactly like to the audio device (with possible jitter buffer >>> under/overrun and playback-speed adjustments) or will the voice sample >>> be written just one after the other to the wave file? >> >> Call audio will be saved one after another into the same file, however, >> this can also be solved in order to receive recording of a single call. > > That's not what I asked - maybe I should make myself more clear: > > During a normal phone call, the receiver may do manipulations to the audio > stream before playing back the audio to the user. For example SIP clients > often have dynamical jitter buffer - when the buffer gets empty the > playback speed will be reduced, when the buffer gets full the playback > speed will be increased, old packets may be ignored completely. > > When a call is recorded, this manipulations are not needed because it > doesn't matter if packets arrive late as for recording there are no > real-time constraints. The receiver can wait until the RTP packets are > received and then it saves the audio payload without need for manipulation > to the wav file. > > If a client writes a wav file like described above, and the is no packet > loss - the recorded file will always be identical to the file sent by the > other party. > > If a client does audio manipulation also for recordings, then the recorded > file will differ from the original and MOS should be different. > > regards > Klaus >
