Theoretical evidence for lack of correspondence between spectral ripple parameters and output of the cochlear implant processor, and an alternate explanation for the correlation with speech perception scores

Abstract

The spectral ripple test is a popular measure of spectral resolution that, in cochlear implant (CI) listeners, has been shown to correlate with speech recognition scores. In the test, listeners distinguish between sounds whose broadband spectra contain some variable number of peaks (i.e. spectral density), with some spectral modulation depth. A meta-analysis of literature on spectral ripple tests in CI listeners shows a generally consistent ceiling level of performance around 2 ripples per octave, with some outliers performing above. We propose that there is a critical point in spectral density at which the output of the CI transitions from representing spectral ripples to a qualitatively different signal, resulting from the interaction between the spectral peaks and the number of active device channels. Artefactual nonlinearities for ripples of higher densities generally do not resemble ripples but may accidentally match the spectral envelope characteristics of speech sounds. The unintended correspondenc...