Speech quality evaluation of artificial bandwidth extension: comparing subjective judgments and instrumental predictions

Artificial bandwidth extension (ABE) methods have been developed to enhance the quality and intelligibility of bandlimited speech transmitted over a telephone connection. Subjective listening tests are the most reliable way of evaluating the quality of ABE, but listening tests are time-consuming and expensive to arrange. Instrumental measures have also been used to estimate the subjective quality of ABE. This study extends the results of an earlier subjective evaluation of ABE methods by instrumental quality predictions computed with WB-PESQ (ITU-T Recommendation P.862.2) and POLQA (ITU-T Recommendation P.863). The instrumental quality predictions are compared with the subjective quality scores. The results indicate that POLQA correlates better with the subjective quality than WB-PESQ. Neither WB-PESQ nor POLQA can predict the rank order of the evaluated ABE methods in all conditions.