This study shows that PESQ can be used as a tool to evaluate degradations from listener echo and duplex impairments caused by echo-mitigation algorithms such as echo cancellation or echo suppression. Both the PESQ-based metric and standards 3GPP TS 26.132 and P.502 share the approach of testing with real speech and comparing an impaired signal to an unimpaired reference. However, unlike 3GPP/P.502, PESQ provides tools for accurate time alignment of the signals that function even with temporally varying delay (jitter) and thus allow measurement in IP-based networks. Moreover, the PESQ metric follows the common practice of calculating PESQ values for any test condition with several speech samples, which stabilizes the quality estimate. In contrast 3GPP prescribes the use of a single test signal, which causes potentially misleading sampling error. Finally, the well-developed perceptual model underlying PESQ generates a perceptually relevant one-dimensional result. This is suitable for benchmark or regression testing. In contrast 3GPP and P.502 use only rudimentary perceptual models or no models at all and generate multidimensional results that are unwieldy when used for performance comparison or tracking
Affiliation: Dolby Laboratories, San Francisco, CA, USA
JAES Volume 67 Issue 3 pp. 124-134; March 2019
Publication Date: February 27, 2019
No AES members have commented on this report yet.
If you are not yet an AES member and have something important to say about this report then we urge you to join the AES today and make your voice heard. You can join online today by clicking here.