Objective Quality Evaluation of Speech Communication Channels Using Dynamic Spectral Features

In this paper an improved objective speech quality evaluation measure is proposed that is based on a combination of (averaged) instantaneous and dynamic spectral features. This measure is based on the average distance measure between subsequent LPC-based cepstral envelopes (LPC cepstral distance) of the original and the distorted utterances. In the current method, the test words are represented by time sequences of LPC cepstrum coefficients and energy, as well as by regression coefficients, being the dynamic measure. The effect of various representative distortions in speech communication channels, such as noise masking, band pass filtering, echo, and peak clipping, on 5 nonsense words from one female persian speaker was measured for objective quality evaluation. Also a subjective quality evaluation was done on these distorted words by using the mean opinion scores (MOS) from 18 persian listeners. The correspondence between this subjective quality measure and the new objective measure appears to be very high (R = 0.954).

کلیدواژه ها:

objective test ، cepstrum ، lpc analysis Subjective test ، distortion