Robust speech recognition in noisy backgrounds based on Teager energy operator and auditory process

In this paper we present a new approach based on Teager energy operator and the properties of the peripheral auditory system for robust speech recognition in noisy environments. The speech signal is first divided into critical bands, and then the Teager energies of each sub-band are estimated. The spectral transformations including intensity to loudness conversion and lateral inhibition are also incorporated according to the human auditory process. Finally, the feature vectors can be constructed by linear predictive (LP) analysis. A speaker-independent Mandarin digits recognition task is performed for evaluating the performance of the proposed front-end. The results show an improved, recognition performance compared to the conventional front-ends such as MFCC and PLP.