A neuronal network model for context-dependence of pitch change perception

Abstract

Many natural stimuli have perceptual ambiguities that can be cognitively resolved by the surrounding context. In audition, preceding context can bias the perception of speech and non-speech stimuli. Here, we develop a neuronal network model that can account for how context affects the perception of pitch change between a pair of successive complex tones. We focus especially on an ambiguous comparison—listeners experience opposite percepts (either ascending or descending) for an ambiguous tone pair depending on the spectral location of preceding context tones. We developed a recurrent, firing-rate network model, which detects frequency-change-direction of successively played stimuli and successfully accounts for the context-dependent perception demonstrated in behavioral experiments. The model consists of two tonotopically organized, excitatory populations, Eup and Edown, that respond preferentially to ascending or descending stimuli in pitch, respectively. These preferences are generated by an inhibitory population that provides inhibition asymmetric in frequency to the two populations; context dependence arises from slow facilitation of inhibition. We show that contextual influence depends on the spectral distribution of preceding tones and the tuning width of inhibitory neurons. Further, we demonstrate, using phase-space analysis, how the facilitated inhibition from previous stimuli and the waning inhibition from the just-preceding tone shape the competition between the Eup and Edown populations. In sum, our model accounts for contextual influences on the pitch change perception of an ambiguous tone pair by introducing a novel decoding strategy based on direction-selective units. The model's network architecture and slow facilitating inhibition emerge as predictions of neuronal mechanisms for these perceptual dynamics. Since the model structure does not depend on the specific stimuli, we show that it generalizes to other contextual effects and stimulus types.

Pitch Class and Envelope Effects in the Tritone Paradox Are Mediated by Differently Pronounced Frequency Preference Regions

“…A further limitation of the threshold model is that it is limited to tone pairs, ignoring the context effect shown in the literature (Englitz et al, 2013 ; Chambers and Pressnitzer, 2014 ; Chambers et al, 2017 ). The neuronal network model by Huang et al ( 2015 ) predicts these context effects and, additionally, the sigmoid response pattern as a function of the pitch classes of tritone pairs without context. It is assumed that different Shepard tones correspond to different presynaptic strengths, affecting a synaptic weighting function.…”

“…The threshold model predicted a variety of ascending/descending patterns for tritone pairs, including patterns of pronounced pitch class and envelope effects and even more complex patterns, depending on the form of the threshold function. The threshold function can be interpreted as an internal spectral weighting function (Terhardt et al, 1982b ) or a synaptic weighting function (Huang et al, 2015 ), with the result that a specific frequency region is particularly important for pitch. When such a preference region is clearly distinct, implemented through steep decreasing threshold functions, the threshold model predicts the typical findings of Deutsch ( 1991 ), that is, pronounced pitch class and small envelope effects.…”

“…Recently, studies have shown that preceding tone sequences also caused adaptation (Dawe et al, 1998 ; Malek and Sperschneider, 2018 ) and other context effects in tritone pairs (Englitz et al, 2013 ; Chambers and Pressnitzer, 2014 ; Chambers et al, 2017 ). Two computational models were recently published to explain these context effects (Huang et al, 2015 ; Chambers et al, 2017 ).…”