Voice Quality Tests (continued)

Perceptual Analysis Measurement System (PAMS)

Developed by British Telecommunications ~1998

ITU-T’s P.563

Passive monitoring
0.85 to 0.9 correlation with human listeners
ITU standard May 2004

Psytechnics algorithm: psyvoip

passive listening
uses RTP statistics

"E Model" - ITU-T G.107

passive monitoring

Transcript

[slide467] There's also Perceptual Analysis Measurement System, PAMS, introduced by British Telecommunications. The ITU's P.563. And it has a little lower correlation, but its big advantage is it's completely passive monitoring. We can just simply collect statistics, and now we can come up with a rating. There's Psytechnics algorithm, also based on passive listening, but it actually uses the RTP statistics. It grabs the data from the RTCP, and says we can look at how many packets were lost, et cetera, and now we can automatically derive what the quality is. And now ITU has the G.107 model, the so-called E-model, that's also based on passive monitoring, and it's a parameterized model that's supposed to be reasonably descriptive from a human listener's point of view, but it's easy to calculate with. But the biggest problem here is if you're going to charge for it, you need to measure it. And it turns out, not only do you need to measure it as an operator, but your customers may even need to measure it. Because they're paying for a given level of quality, they need to make sure that they're getting it. So Ian Marsh, who's a researcher at RISE, ri.se (SiCS), over a nearly decade period of time, made calls between very large numbers of sites, looked at the statistical analysis, and tried to figure out how long do we need to actually collect data before we can predict the quality of the whole session the user would experience. Because we don't have to monitor throughout every session, because that would require a very large amount of monitoring, and it won't scale well. And what he found out is that in fact, within the first few seconds of the call, we can assess what the quality is going to be for the entire session. And most of you have done it yourself. You place a call to somebody, you hear, and you can barely hear the other person, what do you do? You hang up and you make the call again, and hopefully you get a different circuit. It turns out that computers can do the same thing. They can listen for a short period of time to the call, and say is it likely to be a good quality call or not.