SS04: COCHLEAR IMPLANTS AND OTHER AUDITORY IMPLANTS

SIGNAL PROCESSING FOR SPEECH AND MUSIC

Norbert Dillier

Laboratory of Experimental Audiology, ENT Department, University Hospital Zurich, Switzerland

Cochlear implant systems attempt to selectively stimulate small groups of nerve fibers with minimal channel interactions. To achieve this goal, adequate and suitable signal processing strategies had to be developped and appropriate mapping of signal parameters to patient-specific psycho-electrical stimulation conditions  has to be accomplished in order to provide loudness, pitch and timbre percepts which mimic those of normal hearing subjects.

Most coding strategies which are used in today’s CI speech processors divide the input signal into a number of logarithmically spaced frequency bands and generate the stimulus signals based on the signal envelopes in these frequency bands. Depending on the approach, either a preselected fixed number of frequency bands is selected and mapped to a corresponding number of fixed electrodes which are stimulated with a periodic pulsatile stimulus whose amplitude varies according to the envelope fluctuations of the corresponding frequency band and whose stimulation rate is constant (and preferrably as high as possible in order to convey the full range of envelope variations) or a number of frequency bands is selected depending on the energy within each band in decreasing order up to a preselected number.

In all of these strategies, the fine temporal structure which is generated in addition to the envelope fluctuations when a signal is passed through a set of bandpass filters is ignored. This does not seem to be a major problem for the discrimination of different speech sounds. Many implanted subjects reach excellent performance levels when tested with standard open set speech tests. However, most subjects still experience difficulties in acoustically challenging situations such as in reverberant rooms or in the presence of environmental noise or competing talkers. Some of these problems can be alleviated by proper signal preprocessing, directional microphones or induction loop and FM systems. But there seems to be a more fundamental limitation of cochlear implants when it comes to the appreciation of music. The insufficient frequency resolution of a cochlear implant and the lack of preservation of phase information in stimulation patterns are some of the reasons for the poor sound quality perception and weak  performance in music feature discrimination and identification tasks of CI recipients. Current research is seeking new solutions for these challenges.