Harmonic Encoding in Cochlear Implants

Abstract

Today’s standard in cochlear implant (CI) signal processing is based on incoherent tem- poral envelope and temporal fine structure (TFS) extraction. Incoherent envelopes are suffi- cient for the baseline task of speech recognition in quiet; however, current efforts to improve secondary tasks such as speech recognition in noise, lexical tone discrimination and music perception are fundamentally limited by this processing. Harmonic signals are ubiquitous in speech and music. This thesis argues for the ben- efits of coherent extraction of harmonic envelopes and temporal fine structure. By taking harmonic structure into account when designing an envelope and TFS extraction system, processing artifacts can be minimized and signals can be represented more efficiently with the limited data rate of cochlear implants. Furthermore, the proposed method will open up more possibilities for improved cochlear implant encoding. This thesis is a guide to developing a coherent extraction strategy. Incoherent and co- herent extraction systems are evaluated and a generalized method is defined. This method is then applied to harmonic signal encoding. Performance metrics are defined and evaluated and best designs are suggested.