Download files
Abstract
Cochlear implant systems need Automatic Gain Control (AGC) to compress the large dynamic range (~120 dB) of the acoustic environment into the small dynamic range (< 20 dB) of electrical stimulation. This thesis is concerned with the design, implementation and evaluation of AGC systems for cochlear implants. It investigated the effects of AGC on the speech intelligibility of cochlear implant recipients. Various AGC configurations were evaluated with sentences presented over a wide range of levels at different Signal-to-Noise Ratios (SNR) to identify important factors affecting the performance. Signal metrics were developed to quantify the effects of AGC on the channel envelopes. The goal was to improve speech intelligibility in adverse listening conditions.
The performance-intensity functions of cochlear implant recipients with no AGC and with a front-end compression limiter were measured in noise. With no AGC, the proportion of envelope clipping grew monotonically with presentation level. The front-end limiter substantially reduced envelope clipping yet gave little improvement in speech intelligibility. The recipients were highly tolerant of envelope clipping when the background noise was low. SNR degradation was identified as the main factor reducing speech intelligibility.
A front-end limiter cannot guarantee zero envelope clipping. In contrast, the proposed envelope profile limiter eliminated envelope clipping and hence preserved the spectral profile. The two AGCs were evaluated, with two release times (75 and 625 ms). The shorter release time gave worse speech intelligibility because it caused more waveform distortion and output SNR reduction. For a given release time, preserving spectral envelope profile gave additional benefits. In a take-home experiment, cochlear implant recipients rated a program with the envelope profile limiter equivalent to their everyday program.
A conventional cochlear implant signal path uses a predetermined input dynamic range, which is shifted up or down by the AGC. In contrast, the proposed Adaptive Loudness Growth Function (ALGF) continually optimized the input dynamic range by estimating the noise floor and peak level in each channel. The ALGF gave better Speech Reception Threshold (SRT) than the existing state-of-the-art AGC system at the high presentation level when evaluated with a newly developed roving-level SRT test at three presentation levels.