Spectral slope based distortion measures for all-pole models of speech

Previous research on spectral slope distortion measures has considered their use with spectra derived from critical-band filterbank analysis. This paper discusses the approximation and use of spectral slope measures derived from the spectra of all-pole models of speech. Application of these measures to both standard linear predictive (LP) analysis and a recently proposed perceptually-based LP technique are considered. Initial testing of these slope measures for speaker-dependent isolated word recognition indicates that they are quite robust, giving considerable performance improvement over the standard cepstral distance measure in several noisy speech situations.