PESQ (Perceptual Evaluation of Speech Quality) is an objective test measure established by the ITU-T P.862. The PESQ attempts to estimate the mean opinion score (MOS) of a given audio signal. True MOS are obtained through subjective listening tests conducted by trained participants. The utilization of PESQ for the evaluation of algorithms such as Acoustic Echo Cancellation is critical to efficient development. The number of tests that should be performed to fully exhaust the various test permutations make subjective testing unpractical.
PESQ can be used to evaluate the performance of the AEC for the near-end and double-talk scenarios. PESQ is a full-reference test algorithm. In other words, the test requires access to the original clean source signal. For AEC testing, this is the near-end reference speech signal. This signal is then compared with the signal under test. To evaluate the performance of the AEC algorithm, the PESQ test need to be performed twice. Once on the unprocessed microphone signal and again on the AEC processed result.
These tests will confirm that the AEC algorithm does not degrade the PESQ score during near-end only scenarios. In addition, the PESQ score should have a significant improvement during double-talk scenarios, as the echo signal shall no longer be degrading the signal quality. Once the scores have been collected, they can serve as a baseline for regression testing when improvements are made to the AEC algorithm. To help capture the whole picture of the AEC’s performance, the echo return loss enhancement needs to be estimated during echo-only scenarios.