

Author: Marzinzik Mark Kollmeier Birger
Publisher: S. Hirzel Verlag
ISSN: 0001-7884
Source: Acta Acustica, Vol.89, Iss.3, 2003-05, pp. : 521-529
Disclaimer: Any content in publications that violate the sovereignty, the constitution or regulations of the PRC is not accepted or approved by CNPIEC.
Abstract
Most noise reduction schemes proposed in the literature so far that aim at enhancing speech in a noisy background were evaluated using objective quality measures. Since listener tests were mostly not reported, conclusions might be invalid due to the inappropriateness of some objective measures for the specific kind of distortions introduced by noise reduction algorithms. Moreover, if the noise reduction schemes are intended for use in digital hearing aids, the objective measures have to be validated with quality judgments from hearing-impaired subjects. In the present study, different objective speech quality measures were applied to the same test signals that were judged by six hearing-impaired subjects. Single-microphone noise reduction algorithms proposed by Ephraim and Malah (1984, 1985) and a binaural noise reduction algorithm (directional filter and dereverberation) proposed by Wittkop (2001) were employed. The paired comparisons carried out by the subjects were analyzed with the Bradley-Terry model [4] resulting in difference scale values for each algorithm. The Log-Area-Ratio (LAR) objective measure shows the highest correlation with overall subjective preference, while the PMF measure [5] corresponds best with the subjectively perceived amount of noise suppression. A potential application of objective measures is the optimization of noise reduction algorithms by effectively evaluating the often large parameter space.
Related content







