Tailored MFCCs for sound environment classification in hearing aids

Abstract

Hearing aids have to work at low clock rates in order to minimize the power consumption and maximize battery life. The implementation of signal processing techniques on hearing aids is strongly constrained by the small number of instructions per second to implement the algorithms in the digital signal processor the hearing aid is based on. In this respect, the objective of this paper is the proposal of a set of approximations in order to optimize the implementation of standard Mel Frequency Cepstral Coefficient based sound environment classifiers in real hearing aids. After a theoretical analysis of these coefficients and a set of experiments under different classification schemes, we demonstrate that the suppression of the Discrete Cosine Transform from the feature extraction process is suitable, since its use does not suppose an improvement in terms of error rate, and it supposes a high computational load. Furthermore, the use of the most significative bit instead of the logarithm also supposes a considerable reduction in the computational load while obtaining comparable results in terms of error rate.