Performance evaluation of objective quality measures for coded speech

Speech coding is a basic technique that is indispensable to multimedia and personal communications, which require digital coding of the speech signal and information compression. Up to the present, the quality of the speech c oding method has been examined by "subjective measures" where test subjects examine the quality of the coded speech. Since, however, subjective measur es require tremendous time and effort, it is desired to develop "objective measures" that estimate the subjective quality based on the physical featur es of the coded speech. There have been proposed several methods to provide objective measures for the coded speech, but it has been difficult to comp are their performance since there is no common performance evaluation metho d. This article first proposes a performance evaluation method for objectiv e measures, which is then used to compare the performances of typical objec tive measures. It is shown that, under certain conditions, the subjective q uality can be estimated by objective measures with the same accuracy as tha t of subjective measures. In other words, it is shown that the proposed met hod is practically useful. (C) 2000 Scripta Technica.