Improving Speech Intelligibility in Cochlear Implants using Acoustic Models

Cochlear implant (CI) is a prosthetic device that partially replaces the functions of the human ear via electrical stimulation. Cochlear implants are system and/or patient specific that mandates a simulation model prior to implantation. In the present work to improve the perceptual quality of the speech generated by a CI model, system specific parameters are analyzed by developing uniform bandwidth filterbank-based acoustic CI models, an auditory model-based CI system with frequency bands spacing similar to the critical-bands of an auditory system and Mel-frequency cepstral coefficients (MFCC) based analysis-synthesis system for cochlear implants. Acoustic CI simulations are generated for all the vowels of English language and words (easy and hard) from the Lexical Neighbourhood Test (LNT) and sentences from TIMIT database using waveform and feature extraction strategies. A closed-set listening test is conducted and a comparative study is made among the various acoustic CI models developed. The perceptual quality/speech intelligibility of the speech is rated in 5 point grading. It is observed that the acoustic CI simulation for sentences generated by critical-band-based CI system showed a mean opinion score of 4.1 as opposed to 3.1 for uniform bandwidth filters-based CI system. Key-Words: Cochlear implants, Filter banks, Critical band, Speech intelligibility, MFCC, Channel vocoder, Auditory model, Acoustic CI simulations.