Many tools on the market still rely on the e-model for voice assessment, measuring network conditions such as latency and jitter, and running them through an algorithm to generate an R-Factor score.
This score can be translated to a MOS (Mean Opinion Score) value, but MOS-based assessments are fundamentally flawed for use with high-definition VOIP (voice over IP), as they are essentially a comparison to toll-quality voice. High-definition VOIP, with its increased voice sampling capabilities, should automatically exceed these parameters.
Many vendors have adjusted their tools with an updated e-model that considers additional characteristics particular to wideband audio, giving wideband codecs the possibility for higher R-Factor scores. But the translation to MOS values remains unchanged, so users must take care not to compare apples to oranges. A MOS score derived from a wideband call is not directly comparable to one obtained from a narrowband call.
Should you heed the call for high-definition voice? Find out here.
For instance, during tests, I used Network Instruments Observer 12 to do a side-by-side comparison of two calls. I measured the same two phones by first placing a call using the wideband codec G.722 and then using G.711 µ-law. While those I called during the test reported that the wideband codec call sounded much clearer and richer, the voice assessment scores for the wideband call were significantly lower than for the narrowband call.
While our assessment tool helped because it displayed the root cause of the discrepancy—excessive jitter that was endemic to our wideband-enabled network (needing further investigation)—the MOS scores generated were not useful as a comparative measure. But products may react differently to wideband codecs, so administrators should take care to work with the vendor to determine exactly how wideband audio is handled.
There are other voice assessment systems out there that may be able to provide more accurate voice quality assessments, but these systems have had to be adjusted for wideband audio support as well. For instance, PESQ (Perceptual Evaluation of Speech Quality) is an active, reference model-based approach requiring the insertion of a known signal into a voice stream that can be accurately compared with what comes out at the other end of the call. PESQ has the added benefit of not caring what the network transport medium was. But, again, accounting for codecs will be paramount as wideband and narrowband assessments should not be compared directly.
Check out eWEEK.coms for the latest news, views and analysis on voice over IP and telephony.