An improved switch speech enhancement algorithm for automatic speech recognition

In order to improve the performance and save the computational cost of automatic speech recognition (ASR) system in noisy environment, this paper studies a new switch speech enhancement algorithm. Firstly, based on an optimal modified log-spectral amplitude (OM-LSA) estimator and employing an improved a priori signal-to-noise ratio (SNR) estimate based on temporal cepstrum smoothing (TCS), the proposed new approach gains improved noise reduction performance. Next, this new approach and the conventional Wiener algorithm are combined to develop a switch algorithm, whose switching mechanism is designed by the maximum likelihood (ML) estimate of an a priori SNR and voice activity detection (VAD) technology. Both performance improvement and computational cost reduction can be achieved by the proposed switch algorithm. Computer simulations with ASR system verify the effectiveness of the proposed algorithm.