A New Method of Robust Speech Recognition under Noisy Environments

A new robust speech recognition method combining adaptive filtering based on a varibile-step LMS algorithm and perceptual linear prediction cepstral estimation is proposed. The variable-step LMS algorithm is realized by optimizing the non-linear relation between step size and mean square error. The perceptual linear prediction cepstral extracted from the enhanced speech is used for generating the speech feature parameters. The experimental results show that the performance of this robust speech recognition method is better than that of the speech recognition method based on Sigmind function under lower signal-to-noise ratio.