ROOT CEPSTRAL ANALYSIS - A UNIFIED VIEW - APPLICATION TO SPEECH PROCESSING IN CAR NOISE ENVIRONMENTS

The performance of speech recognition systems is significantly degrade d in the presence of noise. To solve the noise problem, there is a nee d to reconsider standard approaches by taking into account this new co nstraint. We first envisage two well-known cepstral representations (p arametric and non-parametric) of speech signals and propose a unifying view of both schemes. We introduce a pseudo-autocorrelation domain, w hich can be interpreted as a ''Root-cepstral domain'', and we show how non-parametric cepstral and linear predictive analyses converge to th e same optimal solution. Experiments are carried out using an HMM-base d isolated word recogniser for speaker-dependent and speaker-independe nt tasks in car noise environments.