Date of Award
Doctor of Philosophy
Cochlear Implants, Hearing in Noisy Conditions, Improving the Intelligibility, Signal Processing
In this thesis, we focus on improving the intelligibility of speech for cochlear implants (CI) users. As an auditory prosthetic device, CI can restore hearing sensations for most patients with profound hearing loss in both ears in a quiet background. However, CI users still have serious problems in understanding speech in noisy and reverberant environments. Also, bandwidth limitation, missing temporal fine structures, and reduced spectral resolution due to a limited number of electrodes are other factors that raise the difficulty of hearing in noisy conditions for CI users, regardless of the type of noise. To mitigate these difficulties for CI listener, we investigate several contributing factors such as the effects of low harmonics on tone identification in natural and vocoded speech, the contribution of matched envelope dynamic range to the binaural benefits and contribution of low-frequency harmonics to tone identification in quiet and six-talker babble background. These results revealed several promising methods for improving speech intelligibility for CI patients. In addition, we investigate the benefits of voice conversion in improving speech intelligibility for CI users, which was motivated by an earlier study showing that familiarity with a talker’s voice can improve understanding of the conversation. Research has shown that when adults are familiar with someone’s voice, they can more accurately – and even more quickly – process and understand what the person is saying. This theory identified as the “familiar talker advantage” was our motivation to examine its effect on CI patients using voice conversion technique. In the present research, we propose a new method based on multi-channel voice conversion to improve the intelligibility of transformed speeches for CI patients.
Azimi, Behnam, "Improving the Speech Intelligibility By Cochlear Implant Users" (2016). Theses and Dissertations. 1345.