A model of auditory deviance detection

Abstract

A key component in computational analysis of the auditory environment is the detection of novel sounds in the scene. Deviance detection aids in the segmentation of auditory objects and is also the basis of bottom-up auditory saliency, which is crucial in directing attention to relevant events. There is growing evidence that deviance detection is executed in the brain through mapping of the temporal regularities in the acoustic scene. The violation of these regularities is reflected as mismatch negativity (MMN), a signature electrical response observed using electro-encephalograpy (EEG) or magneto-encephalograpy (MEG). While numerous experimental results have quantified the properties of this MMN response, there have been few attempts at developing general computational frameworks of MMN that can be integrated in comprehensive models of scene analysis. In this work, we interpret the underlying mechanism of the MMN response as a Kalman-filter formulation that provides a recursive prediction of sound features based on the past sensory information; eliciting an MMN when predictions are violated. The model operates in a high-dimensional space, mimicking the rich set of features that underlie sound encoding up the level of auditory cortex. We test the proposed scheme on a variety of simple oddball paradigms adapted to various features of sounds: Pitch, intensity, direction, and inter-stimulus interval. Our model successfully finds the deviant onset times when the deviant varies from the standard in one or more of the calculated dimensions. Our results not only lay a foundation for modeling more complex elicitations of MMN, but also provide a versatile and robust mechanism for outlier detection in temporal signals and ultimately parsing of auditory scenes.

Mismatch negativity (MMN) to pitch change is susceptible to order-dependent bias

“…In summary, the present study replicated two elements of bias affecting MMN: (1) MMN to the sound first encountered as a deviant is more susceptible to modulation by sequence stability (i.e., the fast vs. slow sequence effects) than MMN to the sound that is first encountered as standard, and (2) within a slowly changing sequence, MMN to the first deviant is larger than to the second deviant. Both of these observations provide clear evidence that there are factors influencing MMN size beyond those accounted for in existing models of the underlying processes (Näätänen, 1984 , 1990 , 1992 ; Javitt et al, 1996 ; Schröger, 1997 ; May et al, 1999 ; Winkler, 2007 ; Garrido et al, 2009 ; Winkler et al, 2009 ; May and Tiitinen, 2010 ; Garagnani and Pulvermüller, 2011 ; Näätänen et al, 2011 ; Wacongne et al, 2012 ; Kaya and Elhilali, 2013 ; Schröger et al, 2013 ). In Introduction, we hypothesized that differences between processing tone frequency and duration (specifically SSA) may result in differences in the order-driven bias effect for these two features.…”

Auditory Event-Related Potentials

“…A predictive coding account of MMN has also been modelled at a more abstract level using a Kalman filtering (Kalman 1960) approach (Kaya and Elhilali 2013). In this case the timing of events is modelled using a separate filter from the one used to model feature distributions.…”

“…However, so far they have either only been implemented at a rather abstract level; e.g. (Garrido et al 2009;Lieder et al 2013;Kaya & Elhilali, 2013), focus exclusively on a single mechanism for explaining MMN; e.g. (May and Tiitinen 2001;Wacongne et al 2012) or account only for MMN responses to unexpected within-event properties Pulvermüller 2008, 2011).…”

Abnormality detection in noisy biosignals

“…Details of the implementation are presented in [8] but are summarized here. The system matrices are assumed to be constant (2).…”

“…In the latter case, the mechanism provides a guided search method: Rather than blindly searching the entire signal for signs of disease, the problem is reduced to classifying what kind of abnormal event has occurred. We adopt a similar framework to one used to model auditory deviance detection processes in the brain [8], and apply this deviance detection scheme to find abnormalities in auscultation signals applied to lung sounds. The proposed scheme employs recursive tracking of temporal patterns in the signal using Kalman filtering, a popular choice in many medical applications [9].…”