Noise compensation methods for hidden Markov model speech recognition in adverse environments

Abstract

Several noise compensation schemes for speech recognition in impulsive and nonimpulsive noise are considered. The noise compensation schemes are spectral subtraction, HMM-based Wiener filters, noise-adaptive HMM's, and a front-end impulsive noise removal. The use of the cepstral-time matrix as an improved speech feature set is explored, and the noise compensation methods are extended for use with cepstral-time features. Experimental evaluations, on a spoken digit database, in the presence of car noise, helicopter noise, and impulsive noise, demonstrate that the noise compensation methods achieve substantial improvement in recognition across a wide range of signal-to-noise ratios. The results also show that the cepstral-time matrix is more robust than a vector of identical size, which is composed of a combination of cepstral and differential cepstral features.

Estimating acoustic speech features in low signal-to-noise ratios using a statistical framework

“…This problem is encountered in automatic speech recognition where noise and speaker mismatches increase word error rates (WERs) (Vaseghi and Milner, 1997;Chung and Hansen, 2013). Matched training and testing gives substantial reductions in WER but is not practical in changing conditions.…”

Uncertainty in Signal Estimation and Stochastic Weighted Viterbi Algorithm: A Unified Framework to Address Robustness in Speech Recognition and Speaker Verification

“…Also, the differences TEER SD (18dB) -TEER SD (0dB) and TEER SI (18dB) -TEER SI (0dB) were dramatically improved by the weighted Viterbi algorithm in combination with the additive noise model. Finally, it is worth mentioning that the performance of SS is highly dependent on the parameters related to the thresholds (Berouti et al, 1979;Vaseghi & Milner, 1997) that are defined to make the technique work properly. In the case of the SS as defined in (9), parameter β , which defines the lower bound for the estimated signal energy, was not optimized for each SNR although its optimum values is case dependent.…”

Efficient Compensation of Spectral Tilt for Speech Recognition in Noisy Environment

“…잡 음이 섞인 음성에서 잡음을 필터링하거나 깨끗한 음성의 파라미터를 추정하는 방법에는 Wiener 필터링 [5] , Kalman 필 터링 [6] , 스펙트럼 차감(spectral subtraction) [7] 및 켑스트 럼 평균 차감(CMS: cepstral mean subtraction) [8] …”