Meta adaption at the auditory-nerve level and its implications for speech intelligibility

Speech intelligibility relies on the faithful coding of the temporal modulations that carry speech information. For such coding to be possible, the auditory periphery must adapt to changes in contextual sound level because the dynamic range of auditory nerve (AN) fibres is limited to ~30 dB, as found in conventional (adapted-rate) measures of rate-level functions (RLFs). Such meta adaptation is defined as the shift of RLFs to higher probe levels as contextual level increases; it was found in small mammals at the AN and IC levels (Dean et al., 2005; Wen et al., 2009).

A first, modelling study compared predictions by a computer model of the auditory periphery (MAP) to previous modelling of meta adaptation at the AN level. While previous studies found the need for phenomenological modelling of meta adaptation at the inner hair-cell (IHC)-AN synapse level, the physiologically-inspired MAP model readily accounts for it. RLF predictions were extracted from the modelled response of 500-AN fibres over a 50 ms test period and immediately following a 400 ms precursor that set context level. Test and context levels were independently varied in the 0-100 dB range for tone pips or noise bursts. In the physiologically-tested 40-80 dB context-level range, meta adaptation in excess of 0.5 dB/dB was predicted when both medial olivocochlear (MOC) and acoustic reflexes were enabled. Disabling the acoustic reflex reduced it to ~0.25 dB/dB. Disabling also the MOC reflex reduced it a little further, but did not remove it. Firing rates measured during a meta-adaptation paradigm are inherently not adapted rates: the 50 ms test-burst duration is so short that the measurement encompasses a fast stimulus-driven mechanism of neurotransmitter release in the cleft with a slower IHC vesicle-store replenishment mechanism and the early part of the slower modulation of basilar membrane vibrations by efferent reflexes. Anaesthesia used in small-mammal electrophysiology is often assumed to suppress efferent reflexes. Therefore, existing measures of meta adaptation do not reveal its full potential. Modelling efferent reflexes demonstrates the potential for normal-hearing meta adaptation reaching close to an ideal 1 dB/dB.

A second, psychophysical study measured speech intelligibility in noise with normal-hearing listeners attending to vocoded speech based on MAP-predicted AN firing patterns. Incorporating MOC and acoustic reflex information in the vocoder significantly improved speech intelligibility. This finding supports a key role of full meta adaptation in the optimal neural coding of speech modulations in quiet or in noise.