A Speech Processing Strategy Based on Sinusoidal Analysis/Synthesis Model for Cochlear Implant User

Background:

In sinusoidal analysis/synthesis model (SM), complex speech signal presented by a few sinusoidal components is perceivable by normal hearing (NH) listeners. Considering the current speech processing strategy in cochlear implants (CIs) and limited number of channels available, this SM would likely to spare out some representation of speech for better perceptual outcomes in CI system.

Objective:

This study compares behavioral and neurophysiological responses to SM speech between NH listeners and CI users by limiting the number of sinusoids and bandwidth of the input speech.

Methods:

Both NH listeners and CI users were to ask recognize the speech sentences from Azbio Tests processed by SM with input bandwidth (4) limited at 1 kHz, 1.5 kHz, 3 kHz, and 6 kHz and number (3) of extracted sinusoidal components limited at 1, 3, and 6. Azbio sentences processed under these (4X3=12) conditions were presented diotically in 100 repetition to NH listeners, while EEG is recorded. EEG responses were recorded on the same NH listeners with the stimuli noise-vocoded.

Results:

The word recognition scores of both NH listeners and CI users decreases systematically when input bandwidth and sinusoidal components decreases. Average EEG responses evoked by the sentence in NH listeners were correlated to those evoked by their noise-vocoded counterparts.

Conclusions:

Reasonable word scores with NH listeners and CI users were maintained when input bandwidth is 3 kHz and sinusoidal components as little as 3. However, more physiological data with CI users are necessary to derive the relevant parameters in SM for CI system.