Voice quality standards

The legacy voice quality standard is the Mean Opinion Score. A MOS is derived by having people rate the quality of test sentences read aloud over the same communications circuit by male and female speakers. Each sentence is assigned the score of (1) bad, (2) poor, (3) fair, (4) good or (5) excellent, and the scores are averaged.

This expensive, labor-intensive process is reserved for laboratory settings, with no real-time applicability. The MOS scoring is also highly subjective; different listeners might rate the same sentence at opposite ends of the scale.

The industry turned to computers for consistency and economy, and the Perceptual Evaluation of Speech Quality (PESQ) algorithm was co-developed by KPN/TNO and Psytechnics and formalized as ITU P.862. PESQ measures the distortion of test voice signals as they move through a VoIP network and estimates a MOS. While this works fairly well, PESQ sometimes gives high marks to poor signals, and low marks to conversations that sound just fine to human listeners.

To continue reading, register here and become an Insider. You'll get free access to premium content from CIO, Computerworld, CSO, InfoWorld, and Network World. See more Insider content or sign in.

The legacy voice quality standard is the Mean Opinion Score. A MOS is derived by having people rate the quality of test sentences read aloud over the same communications circuit by male and female speakers. Each sentence is assigned the score of (1) bad, (2) poor, (3) fair, (4) good or (5) excellent, and the scores are averaged.

This expensive, labor-intensive process is reserved for laboratory settings, with no real-time applicability. The MOS scoring is also highly subjective; different listeners might rate the same sentence at opposite ends of the scale.

The industry turned to computers for consistency and economy, and the Perceptual Evaluation of Speech Quality (PESQ) algorithm was co-developed by KPN/TNO and Psytechnics and formalized as ITU P.862. PESQ measures the distortion of test voice signals as they move through a VoIP network and estimates a MOS. While this works fairly well, PESQ sometimes gives high marks to poor signals, and low marks to conversations that sound just fine to human listeners.

Also, PESQ is for active monitoring, and as such is intrusive. The newer ITU P.563 standard, like Telchemy's proprietary VQmon technology, is for non-intrusive monitoring of voice streams. This passive approach lets call quality be estimated as conversations occur.

The ITU also established G.107, commonly known as the E Model. This model considers the mouth-to-ear characteristics of a given speech path and assigns an R value between 0 to 100, although the practical range for the ubiquitous G.711 codec is 50 to 94. The actual formula is R = Ro - Is - Id - Ie + A, where

•  Ro is a base factor determined from loudness.

•  Is are impairments that occur simultaneously with speech.

•  Id are impairments that are delayed relative to speech.

•  Ie is an equipment impairment factor.

•  A is an advantage factor.

The E Model is a scalable, lightweight protocol for making repeated measurements throughout the call.

Voice quality adds a new dimension to network management. The IETF addressed VoIP management with RTP Control Protocol Extended Reports (RTCP XR), a protocol that can be implemented inexpensively via software in IP phones and gateways. RTCP XR measures call quality using such key metrics as packet loss and discard; delay; signal, noise and echo levels; and configuration data. These metrics can be sampled midstream by a probe or analyzer, or an SNMP system can collect them from a gateway.

Back to feature: "New tools quantify VoIP call quality"

Read more about voip & convergence in Network World's VoIP & Convergence section.