In recent years several measures of distortion between speech waveforms have been proposed as substitutes for the traditional but subjectively inadequate mean-squared error. All of these measures involve some form of distortion measure between the second order properties of the speech processes producing the waveforms instead of an average of the waveform error power. In particular, they depend on the power spectral densities or linear models of the speech process. In this report the properties and interrelations of several such measures are developed. In particular, the relative strengths or equivalences of the various implications and applications of these measures to prediction, detection, and coding are summarized. (Author)
Memory and computation requirements imply fundamental limitations on the quality that can be achieved in vector quantization systems used for speech waveform coding and linear p...
An efficient and intuitive algorithm is presented for the design of vector quantizers based either on a known probabilistic model or on a long training sequence of data. The bas...
Many problems in voice recognition and audio processing involve feature extraction from raw waveforms. The goal of feature extraction is to reduce the dimensionality of the audi...
Four quantization schemes for the reflection coefficients obtained from linear prediction speech analysis are theoretically compared. The asymptotic performance of each scheme i...
This paper studies summary measures of the predictive power of a generalized linear model, paying special attention to a generalization of the multiple correlation coefficient f...