You are here: Home / Research / Bernstein II / Individual Projects / Project C2 – Novel coding strategies for cochlear implants: from fine time structures to sparse neural representations

Project C2 – Novel coding strategies for cochlear implants: from fine time structures to sparse neural representations

Ingeborg Hochmair (MED-EL), Lutz Wiegrebe(LMU), and Werner Hemmert (TUM)

Cochlear implants (CIs) belong to the most successful neural prostheses. Due to the steady progress during the last three decades CIs have become an accepted treatment method for adults and children with profound hearing loss, effectively replacing a full sensory organ. Confidence in technology has led to an extension of the implantation criteria even to patients with partial hearing loss, who benefit from acoustic and electric stimulation via a combined speech processor in MED-EL devices [1]. Technically, the most severe limitations of CIs arise from electrode interactions, both spatially due to cross-talk and in the time domain. Therefore only few independent frequency channels are available and most coding strategies stimulate them in sequential order. As a consequence, CI patients lag behind normal hearing subjects especially in music appreciation, the perception of temporal cues for sound localization, and speech intelligibility in noisy environments.

Objectives and description of the project

We will derive novel coding strategies based on our inner ear model and sparse coding algorithms, which elicit electrical impulses precisely synchronized to the sound signal. The coding strategies will provide temporal information with sufficient precision to enable sound localization. We will evaluate and optimize the coding strategies with our model of electric hearing using the frameworks of information theory [2] and automatic speech recognition [3], before we extensively test them in patients with cochlear implants. When testing temporal cues for sound localization, we expect that the neural decoding mechanisms require precise adjustment [4] (see also project D2). Therefore, in CI patients decoding of spatial cues will probably be de-tuned and require training to re-emerge [4]. Since our strategies will go beyond sequential stimulation of electrodes, we will have to develop novel methods to fit the electrical stimuli to the narrow and rate-dependent dynamic range of the electrically excited auditory nerve.

[1]: Holmberg and Hemmert J Acoust Soc Am submitted, 2009. [2]: Wiegrebe and Meddis J Acoust Soc Am 2004. [3]: Wang et al. IEEE ICASSP 2006. [4]: Holmberg et al. Speech Com 2007.