This paper presents an inversion algorithm for dynamic Bayesian networks towards robust speech recognition, namely DBNI, which is a generalization of hidden Markov model inversion (HMMI). As a dual procedure of expectation maximization (EM)-based model reestimation, DBNI finds the 'uncontaminated' speech by moving the input noisy speech to the Gaussian means under the maximum likelihood (ML) sense given the DBN models trained on clean speech. This algorithm can provide both the expressive advantage from DBN and the noise-removal feature from model inversion. Experiments on the Aurora 2.0 database show that the hidden feature model (a typical DBN for speech recognition) with the DBNI algorithm achieves superior performance in terms of word error rate reduction.

This letter proposes a novel approach for mel-cepstral analysis based on the psychoacoustic model of MPEG. A perceptual weighting function is developed by applying cubic spline interpolation on the signal-to-mask ratios (SMRs) which are obtained from the psychoacoustic model. Experiments on speaker identification and speech re-synthesis showed that the proposed method not only improved the speaker recognition performance, but also improved the speech quality of the re-synthesized speech.