Speech recognizers often experience serious performance degradation when deployed in an unknown acoustic (particularly, noise contaminated) environment. To combat this problem, we proposed in a previous study a family of new distortion measures that were shown to be able to withstand additive white noise without requiring 1) explicit knowledge of the noise, 2) noise reduction provisions, or 3) reference template retraining. One particularly effective distortion memure in the family is the one

more »

... at takes into account the norm shrinkage bias in the noisy cepstrum. In this paper, we incorporate a firstorder equalization mechanism, specifically aiming at avoiding the norm shrinkage problem, in a hidden Markov model (HMM) framework to model the speech cepstral sequence. Such a modeling technique requires special care as the formulation inevitably involves parameter estimation from a set of data with singular dispersion. We provide solutions to this HMM stochastic modeling problem and give algorithms for estimating the necessary model parameters. We experimentally show that incorporation of the first-order mean equalization model makes the HMM-based speech recognizer robust to noisc:. With respect to a conventional HMM recognizer, this leads to an improvement in recognition performance which is equivalent to about 15-20 dB gain in signal-to-noise ratio.