The contribution of high-frequency information to fine-grained speech perception in cochlear-implant-simulated speech

It has been observed that the contribution of information in the higher frequency ranges (>7000 Hz) to speech perception is negligible. This assumption has influenced cochlear implant processing strategies, which typically filter out high-frequency information. However, even if speech recognition is robust without high-frequency information, that information may nevertheless contribute to the actual processing of speech. We asked 15 normal-hearing participants to categorize words along 8-step b/p (e.g., beach to peach) and s/sh (e.g., self to shelf) continua in normal and CI-simulated speech. The CI simulations used a bandpass filter with 8 bands between 200 and 7000 Hz. The envelope for each band was extracted and used to amplify broadband noise within the band’s frequency range. Results showed that participants’ boundaries shifted significantly for both continua; the endpoints of each continuum were categorized consistently, but ambiguous tokens were more often categorized as sh and b when presented in CI-simulated speech. That is, eliminating high-frequency information during simulation also eliminated information necessary to the perception of sounds with high-frequency cues. These results suggest that while the loss of high-frequency information may not affect CI users’ ability to recognize speech, it may compromise their fine-grained discrimination.