On the application of variable-step adaptive noise cancelling for improving the robustness of speech recognition

As speech recognition and spoken language technologies are being transferred to real applications, the need for greater robustness against adverse noise is becoming increasingly apparent. This paper researches a robust speech recognition method based on adaptive noise cancelling (ANC). It obtained the enhanced speech signal by applying a variable-step adaptive noise cancelling algorithm to reduce noise as pre-treatment of speech recognition under strong noise circumstance. Mel-frequency cepstral coefficients (MFCC) were then computed as recognition features. Compared with conventional spectral subtraction (SS), standard MFCC recognizer and adaptive noise cancelling algorithm in literature, experimental results indicate that this method performs better when signal-to-noise ratio (SNR) ranges from -10 to 15 dB. In addition, the presented method denotes good noise robustness when SNR decreases.