Objective assessment of multichannel cochlear implants: Comparison between different strategies for vowel recognition

摘要

In order to improve and facilitate the assessment of speech recognition performances with cochlear implants (CIs), we have proposed a method [1] based on the processing of large speech databases by the CI and the transfer of the implant output signals on computer for the purpose of automatic classification into phonetic categories (TRANSCLAS method). Automatic assessments obtained with a Digisonic DX10 suggested that correct classification of vowels is better with the "Mel" tuning strategy, in which Digisonic channels are equally spaced on a psychoacoustic scale, than with the "Linear" strategy, in which channel frequencies are equally spaced in Hz. The better performance of the Mel strategy was due to a higher channel density in the vowel formant frequency range. The main purpose of the present work is to see if the implant performance can be improved by a further increase of channel density in the formant frequency range and by using an energy gain of 6 dB per octave ("Mel Modified High Shaping Strategy", or "MMHS"). Automatic assessments suggest that these strategy modifications have a slight negative effect when all 15 channels are active. However, vowel recognition is more resistant to signal distortions with the MMHS strategy. [References: 35]