Intrinsic Noise Improves Speech Recognition in a Computational Model of the Auditory Pathway

Abstract

Noise is generally considered to harm information processing performance. However, in the context of stochastic resonance, noise has been shown to improve signal detection of weak sub- threshold signals, and it has been proposed that the brain might actively exploit this phenomenon. Especially within the auditory system, recent studies suggest that intrinsic noise plays a key role in signal processing and might even correspond to increased spontaneous neuronal firing rates observed in early processing stages of the auditory brain stem and cortex after hearing loss. Here we present a computational model of the auditory pathway based on a deep neural network, trained on speech recognition. We simulate different levels of hearing loss and investigate the effect of intrinsic noise. Remarkably, speech recognition after hearing loss actually improves with additional intrinsic noise. This surprising result indicates that intrinsic noise might not only play a crucial role in human auditory processing, but might even be beneficial for contemporary machine learning approaches.

The Erlangen model of tinnitus development—New perspective and treatment strategy

“…Die Information wird somit immer dann in der zeitlichen Abfolge von Aktionspotenzialen erkannt, wenn die Zeitabstände zwischen den einzelnen Aktionspotenzialen nicht völlig zufällig verteilt sind, sondern die Intervalle der entsprechenden Eigenfrequenz des jeweiligen Frequenzkanals enthalten. Offensichtlich erfordert die Berechnung der Autokorrelation ein präzises Timing der Aktionspotenziale, d. h. eine Phasenkopplung der Aktionspotenziale an die Wellenform oder Umhüllende des Reizes [ 16 ]. Bei Säugetieren ist eine solche Phasenkopplung (im Hörnerv) selbst mit Hilfe des Wever-Salvenprinzips aber nur für Frequenzen bis zu etwa 5 kHz möglich [ 5 , 14 ].…”

Is phase locking crucial to improve hearing thresholds in tinnitus patients?

“…This makes the somatosensory input, receiving mostly information from the neck and pinna muscles, to one of the most likely candidates and would also explain phenomena like somatic tinnitus (Shore et al, 2007). In case of a reduced cochlear input due to inner hair cell damage (Tziridis et al, 2021) or denervation (‘hidden hearing loss’, Kujawa & Liberman, 2009), the added neuronal noise can lift the otherwise subthreshold cochlear input above the threshold (=SR), thereby improving hearing thresholds (Gollnast et al, 2017) or even speech perception (Schilling et al, 2022). For SR to optimize information transmission in the described way, the amplitude of the noise (i.e., the spike rate of the incoming uncorrelated neuronal noise) has to be constantly adapted to changing levels of cochlear input, most probably due to modulation of the inhibition (Chen et al, 2022).…”

“…About one sixth of the general population suffers from chronic, subjective tinnitus (Shargorodsky et al, 2010). In contrast to the common sense that views tinnitus as a maladaptive process that is triggered by some kind of hearing loss, we have put forward a model of tinnitus development that views the phantom percept as a side effect of a mechanism that permanently optimizes information transmission into the auditory system by means of stochastic resonance (SR) (Krauss et al, 2016(Krauss et al, , 2017Schilling et al, 2021Schilling et al, , 2022. In other words, the effect of SR is not limited to pure tone thresholds, which are usually investigated in the 'normal' audiological studies, but seems to be a fundamental principle of (mammal) hearing and therefore auditory perception.…”