Bayesian Feature Enhancement for Reverberation and Noise Robust Speech Recognition

Abstract

Abstract-In this contribution we extend a previously proposed BAYESIAN approach for the enhancement of reverberant logarithmic mel power spectral coefficients for robust automatic speech recognition to the additional compensation of background noise. A recently proposed observation model is employed whose time-variant observation error statistics are obtained as a side product of the inference of the a posteriori probability density function of the clean speech feature vectors. Further a reduction of the computational effort and the memory requirements are achieved by using a recursive formulation of the observation model. The performance of the proposed algorithms is first experimentally studied on a connected digits recognition task with artificially created noisy reverberant data. It is shown that the use of the time-variant observation error model leads to a significant error rate reduction at low signal-to-noise ratios compared to a time-invariant model. Further experiments were conducted on a 5000 word task recorded in a reverberant and noisy environment. A significant word error rate reduction was obtained demonstrating the effectiveness of the approach on real-world data.Index Terms-Robust automatic speech recognition, modelbased BAYESIAN feature enhancement, observation model for reverberant and noisy speech, recursive observation model.

Single-channel dereverberation by feature mapping using cascade neural networks for robust distant speaker identification and speech recognition

“…It can be divided into linear filtering, spectrum enhancement, and feature enhancement. The linear filtering dereverberates time-domain signals or STFT coefficients, e.g., [3,4], the spectrum enhancement dereverberates the corrupted power spectra of signal, e.g., [5][6][7], and the feature enhancement dereverberates the corrupted feature vectors, e.g., [8][9][10]. Meanwhile, the back-end-based approaches attempt to modify the acoustic model and/or decoder so that they are suitable for reverberant environment, e.g., [11,12].…”

“…Several single-channel feature enhancement methods have been proposed. Some of them do not need stereo data at all, e.g., cepstral mean normalization (CMN) [16,17], long-term feature normalization [18], vector Taylor series (VTS) [19], particle filter [8,20], and extended Kalman filter [9,21]. Meanwhile, some of them assume that stereo training data can be acquired.…”