Rhythmic modulation of prediction errors: a possible role for the beta-range in speech processing

Abstract

Natural speech perception requires processing the current acoustic input while keeping in mind the preceding one and predicting the next. This complex computational problem could be handled by a multi timescale hierarchical inferential process that coordinates information flow up and down the language hierarchy. While theta and low-gamma neural frequency scales are convincingly involved in bottom-up syllable-tracking and phoneme-level speech encoding, the beta rhythm is more loosely associated with top-down processes without being assigned yet a specific computational function. Here we tested the hypothesis that the beta rhythm drives the precision of states during the speech recognition hierarchical inference process. We used a predictive coding model that recognizes syllables on-line in natural sentences, in which the precision of prediction errors is rhythmically modulated, resulting in alternating bottom-up vs. top-down processing regimes. We show that recognition performance increases with the rate of precision updates, with an optimal efficacy in the beta range (around 20 Hz). The model further performs when prediction errors pertaining respectively to syllable timing and syllable identity oscillate in antiphase. These results suggest that online syllable recognition globally benefits from the alternation of bottom-up and top-down dominant regime at beta rate, and that the gain is stronger when different features are also analyzed in alternation. These results speak to a discontinuous account of inferential operations in speech processing.

A deep hierarchy of predictions enables online meaning extraction in a computational model of human speech comprehension

“…The discrete portion of our model, or in theory any model with explicit structural and timing information [11,53], can provide a template for organizing distributed oscillatory activities into functional hierarchies through correlating latency-and frequency-specific neuronal dynamics with model-derived information metrics. In general, sensory inputs sampled by fast (gamma) oscillation are parsed into higher-level information as phase alignments of slow (theta, delta) oscillations [26,75,[78][79][80][81], which are found to be modulated by level-specific speech information [32,36,61] and top-down coordination of mid-range (alpha, beta) oscillations [77,78,[82][83][84][85][86]. One promising avenue that exploits both model-derived computational metrics and neural oscillations to disentangle neural information transfer is via a forward model that explains the neurophysiological signal as a result of input-modulated changes in direction-specific connection strengths between specific neural sources (brain areas), i.e., effective connectivity [87,88].…”

A brain-rhythm based computational framework for semantic context and acoustic signal integration in speech processing

“…Although BRyBI shows promising results, it leads to multiple avenues for extensions and improvements through the implementation of more biological mechanisms for rhythm generation, the incorporation of phase-amplitude coupling (PAC) mechanisms, and considering the role of beta in the inference hierarchy [81].…”

“…Following the example of previous similar models [29,60,81], the GM splits each syllable into 8 parts. It allows more flexibility in shaping the auditory spectrogram of syllables and phonemes.…”

A deep hierarchy of predictions enables assignment of semantic roles in online speech comprehension

“…For inter-level information, it is possible to decompose the KLD into bottom-up prediction errors and top-down priors (Friston et al, 2017), and apply known neurophysiological probes, e.g. neural oscillations, to distinguish these two flows in the brain (Bastos et al, 2012, Giraud and Arnal, 2018, Giraud and Poeppel, 2012, Arnal and Giraud, 2012, Bastos et al, 2020, Hovsepyan et al, 2022).…”

“…The discrete portion of our model, or in theory any model with explicit structural and timing information (11,53), can provide a template for organizing distributed oscillatory activities into functional hierarchies through correlating latency-and frequency-specific neuronal dynamics with model-derived information metrics. In general, sensory inputs sampled by fast (gamma) oscillation are parsed into higher-level information as phase alignments of slow (theta, delta) oscillations (26,76,(79)(80)(81)(82), which are found to be modulated by level-specific speech information (32,36,61) and top-down coordination of mid-range (alpha, beta) oscillations (78,79,(83)(84)(85)(86)(87). One promising avenue that exploits both model-derived computational metrics and neural oscillations to disentangle neural information transfer is via a forward model that explains the neurophysiological signal as a result of input-modulated changes in direction-specific connection strengths between specific neural sources (brain areas), i.e.…”