Skip to main content
. Author manuscript; available in PMC: 2022 Jun 26.
Published in final edited form as: Nat Biomed Eng. 2021 May 3;6(6):717–730. doi: 10.1038/s41551-021-00707-y

Fig. 6. Hearing aid compression decreases the selectivity of neural responses to speech.

Fig. 6

a, Spectrograms showing the log power across frequencies at each time point in one instance of the syllable “za” before and after processing with a hearing aid. b, Left, percent change in RMS contrast of all syllables (n = 384 instances with 16 consonants followed by each of 4 vowels spoken by each of 8 talkers) after processing with a hearing aid. Only the first 150 ms of each syllable were used. Right, Performance of a support-vector-machine classifier trained to identify consonants based on spectrograms either before (Original) or after (HA) processing with a hearing aid (mean ± standard error across 10 different held-out samples). c, Left, Performance of a support-vector-machine classifier trained to identify consonants based on population single-trial responses to speech at 62 dB SPL. Results are shown for normal hearing gerbils (NH) and gerbils with hearing loss without a hearing aid (HL), with a hearing aid (HA), and with linear amplification (HL+20dB) (values for each population are shown along with mean ± 95% confidence intervals derived from bootstrap resampling across populations). Right, Magnitude of the differential signal component for single-unit responses to speech at 62 dB SPL (mean ± 95% confidence intervals derived from bootstrap resampling across neurons). d, Performance of a support-vector-machine classifier trained to identify consonants based on population single-trial responses to speech at 62 dB SPL. Results are shown for speech in the presence of ongoing speech from a second talker at equal intensity and speech in the presence of multi-talker babble noise at equal intensity, presented as in c.