Use this URL to cite or link to this record in EThOS:
Title: Modelling the effect of cochlear implant filterbank characteristics on speech perception
Author: Chowdhury, Shibasis
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Restricted access.
Access from Institution:
The characteristics of a cochlear implant (CI) filterbank determine the coding of spectral and temporal information in it. Hence, it is important to optimise the filterbank parameters to achieve optimal benefit in CI users. The present thesis aimed at modelling how the manipulation of the filterbank analysis length and the assignment of spectral channels may effect CI speech perception, using CI acoustical simulation techniques. Investigations were carried out to study the efficacy of providing additional spectral information in low and/or mid frequency channels using a longer filterbank analysis window, with respect to CI processed speech perception in various types of background noise. However, the increase of filterbank analysis length has an associated trade-off, which is a reduction in temporal information. Only a few CI acoustic simulations studies have modelled the characteristics of the FFT filterbank, the most commonly used filterbank in commercial CI processors. An initial experiment was carried out to validate the CI acoustical simulation technique used in the present thesis that implemented an FFT filterbank analysis. Next, the effect of a reduction in temporal information with the increase of the FFT analysis window length was studied. A filterbank with 16 ms analysis window, without the implementation of its finer spectral coding abilities, performed marginally poorer to that of a 4 ms analysis window in a sentence recognition test. The finer spectral coding abilities of the filterbank with 16 ms analysis window, when implemented, revealed that CI processed speech perception in noise could be significantly improved if additional spectral information is provided in the low and mid frequencies. The assignment of additional spectral channels to the low and mid frequencies led to a corresponding reduction in spectral channels assigned to high frequencies. However, no detrimental effect in speech perception was observed as long as at least two spectral channels represented information above 3 kHz. The assignment of additionallow and mid frequency spectral channels also led to significant levels of spectral shift. The significant benefits from additional low and mid frequency information, however, were lost when the effects of spectral shift were introduced in acute experiments, without any training or acclimatisation period. The findings of the present thesis highlight that a longer filterbank analysis, such as 16 ms, may be implemented in CI devices without the fear of any perceptual cost due to a reduction in temporal information, at least for tasks that do not require talker separation. Providing additional low and mid frequency spectral information with a longer filterbank analysis has the potential to improve CI speech perception. However, to obtain potential benefits, the effects of spectral shift should be overcome. The findings of this thesis, however, need to be interpreted considering the limitations of CI acoustical simulation experiments.
Supervisor: Verschuur, Carl Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: RF Otorhinolaryngology