Wavelet Filtering Based on Mellin Transform Dedicated to Cochlear Prostheses

Current speech processing strategies for cochlear prosthesis system use filter bank structure in order to extract speech signal's energies relatively to the multiple frequency bands which are associated to the dedicated stimulation electrodes. This research concerned wavelet filtering module based on Mellin transform and dedicated to one cochlear speech processing strategy. Our bank filtering module was composed of twenty one filters designed according to Morlet mother wavelet and distributed along the Munich critical bands. Hence, our considered auditory spectrum would be divided into twenty one critical frequency bands according to ERB model (equivalent rectangular bandwidth) which would be similar to auditory model. Continuous wavelet transform based filters' outputs would processed in order to provide the relatively energy levels which would be considered as the targeted stimulation levels. Mellin transform used in this processing permitted a fast algorithm execution compared to one standard CWT based filtering module. Algorithm validation was approved using different input signals as pure harmonics first and then real speech signals extracted from TIMIT database. Our proposed strategy ensures not only execution rapidity but also flexibility in programming, easiness in use as well as different safety features that could help reaching each individual's needs.