A Novel Noise-Robust Speech Recognition System Based on Adaptively Enhanced Bark Wavelet MFCC

Bark wavelet is a new wavelet which is especially designed for speech signal. Its base function satisfies time and bandwidth product least. Moreover, the Bark wavelet divides frequency band based on auditory model. In order to enhance the ability to resist the noises of different environments, an adaptive enhancement approach is introduced. With this method, it can also solve the problem of poor understandability of the speech signals. This paper uses Bark wavelet in MFCC. It was used to make preprocessing before FFT. On the other hand, it was used to instead of DCT in MFCC for overcoming the DCT´s disadvantage of fixed time-frequency resolution. Thus, a kind of good anti-noisy speech feature coefficient was obtained. Experimental results of speech recognition demonstrate that this new feature is more robust than the MFCC feature in noise environment and large vocabulary.