Comparison of speech processing strategies for the design of an ultra low-power analog bionic ear

Miniaturizing area and power consumptions of cochlear prosthetic devices is strongly required for full implantation. In this paper, several speech encoding strategies are studied and compared in order to find a compact speech processor that allows for full implantation and is able to convey both time and frequency components of the incoming speech to a set of electrical pulse stimuli. The study covers the widely recognized continuous time interleaved sampling (CIS) and strategies that convey the temporal fine structure (TFS), including race-to-spike asynchronous interleaved sampling (AIS), phase-locking (PL) using zero-crossing detection (ZCD), and PL using a peak-picking (PP) technique. To estimate the performances of the four systems, a spike-based reconstruction algorithm is employed to retrieve the original sounds after being processed by different strategies. The correlation factors between the reconstructed and original signals imply that strategies that convey TFS outperform CIS. Among them, the peak picking technique combines good performance with great compactness since envelope detectors are not required.