SII and fit-to-target analysis of compression system performance as a function of number of compression channels

This work was undertaken to answer the question, 'How does the speech audibility/fit-to-gain-target provided by compression change with number of channels?' For each of 957 audiograms and a given number of compression channels, the channel crossover frequencies were set either to maximize the SII (speech intelligibility index) for low- and high-level speech spectra, or to optimize the fit-to-gain targets from the Cambridge method for loudness equalization (CAMEQ). The audiograms comprised all common configurations, and losses ranged from mild to severe. Use of these computational procedures allowed the predicted, channel-number-based performance to be determined separately from the effects of other compression parameters. From one to five channels were sufficient to yield predicted speech recognition performance within 5% of maximum for 90% of the 'mild' and 'moderate' audiograms. Three to nine channels were necessary for the same level of predicted performance for 90% of the 'severe' audiograms. Four channels or fewer were sufficient to produce less than 5 dB rms error in fit to CAMEQ targets for 90% of all audiograms.