An approach to robust phoneme classification with auditory cortical representation for speech

MARTÍNEZ, C. E.; GODDARD, J. C.; MILONE, D. H.; RUFINER, H. L.

Rosario, Santa Fé, Argentina

Workshop; XIII Reunión de Trabajo en Procesamiento de la Información y Control (RPIC 2009); 2009

In this work, a first approach to a robust speech recognition task by means of a biologically-inspired feature extraction method is presented. The proposed technique provides an approximation to the speech signal representation at the auditory nerve level. It is based on an optimal dictionary of atoms, estimated from auditory spectrograms, and the Matching Pursuit algorithm to obtain the activations. This provides a sparse coding with some basic degree of noise robustness that can be exploited in the application. The recognition task consisted in the classification of a set of 5 highly confusing English phonemes, on clean and noisy conditions. Multilayer perceptrons were trained as phoneme classifiers and the performance was compared to that obtained by a classic parameterization in this task, the Mel frequency cepstral coefficients. Results showed an improvement in the recognition rate using the approximated auditory cortical representation for both the clean case and in the presence of additive white noise.