H Marvi - Centre for Vision, Speech and Signal Processing University of Surrey, Guildford, UK
E Chilton

In previous work the two dimensional root cepstrum (TDRC) has been introduced as a novel method of feature extraction [8]. It has the advantage of an adjustable parameter which can be used to optimise the feature extraction process, reducing the dimensions of the feature matrix and simple computation. In the work described here, the TDRC method has been suggested as a robust technique for noisy speech recognition. Gaussian noise have been added to the all original speech to simulate a noisy environment and to have various of SNR. A task of multi-speaker isolated word recognition is conducted to demonstrated the effectiveness of using these robust features. Experimental results suggest that this method is more robust against noise than original TDC.

Noisy Speech Recognition, Root Cepstrum,Feature extraction, TDRC