Including measures of high gamma power can improve the decoding of natural speech from EEG

Abstract

<jats:title>ABSTRACT</jats:title><jats:p>The human auditory system is adept at extracting information from speech in both single-speaker and multi-speaker situations. This involves neural processing at the rapid temporal scales seen in natural speech. Non-invasive brain imaging (electro-/magnetoencephalography [EEG/MEG]) signatures of such processing have shown that the phase of neural activity below 16 Hz tracks the dynamics of speech, whereas invasive brain imaging (electrocorticography [ECoG]) has shown that such rapid processing is even more strongly reflected in the power of neural activity at high frequencies (around 70-150 Hz; known as high gamma). The aim of this study was to determine if high gamma power in scalp recorded EEG carries useful stimulus-related information, despite its reputation for having a poor signal to noise ratio. Furthermore, we aimed to assess whether any such information might be complementary to that reflected in well-established low frequency EEG indices of speech processing. We used linear regression to investigate speech envelope and attention decoding in EEG at low frequencies, in high gamma power, and in both signals combined. While low frequency speech tracking was evident for almost all subjects as expected, high gamma power also showed robust speech tracking in a minority of subjects. This same pattern was true for attention decoding using a separate group of subjects who undertook a cocktail party attention experiment. For the subjects who showed speech tracking in high gamma power, the spatiotemporal characteristics of that high gamma tracking differed from that of low-frequency EEG. Furthermore, combining the two neural measures led to improved measures of speech tracking for several subjects. Overall, this indicates that high gamma power EEG can carry useful information regarding speech processing and attentional selection in some subjects and combining it with low frequency EEG can improve the mapping between natural speech and the resulting neural responses.</jats:p>

Delta/theta band EEG differentially tracks low and high frequency speech-derived envelopes

“…Specifically, we asked whether these are represented in a similar manner as real speech. In large our data show that this is indeed the case: Envelopes were tracked over fronto-central electrodes at EEG frequencies below 8 Hz, consistent with previous studies (Drennan and Lalor, 2019; Etard and Reichenbach, 2019; Kayser et al, 2015; Mai and Wang, 2019; Synigal et al, 2019). However, the envelopes derived from a higher spectral range (defined here as 2.66 – 8 kHz) were reflected in spatially distinct EEG topographies compared to the envelopes from the lower spectral range (i.e.…”

Auditory stimulus-response modeling with a match-mismatch task

“…A second direction is to improve EEG analysis. Standard models (including those reported here) exploit low-frequency components within the EEG, but useful information may also be carried by high-frequency power (Synigal et al, 2020;Forte et al, 2017;Teoh et al, 2019). If the relevant sources have low SNR, they may not be exploitable without appropriate spatial filtering, but standard linear techniques to find the filters (such as CCA) are not directly applicable.…”

Auditory Stimulus-response Modeling with a Match-Mismatch Task

“…A second direction is to improve EEG analysis. Standard models (including those reported here) exploit low-frequency components within the EEG, but useful information may also be carried by high-frequency power (Synigal et al, 2020; Forte et al, 2017; Teoh et al, 2019). If the relevant sources have low SNR, they may not be exploitable without appropriate spatial filtering, but standard linear techniques to find the filters (such as CCA) are not directly applicable.…”

“…A second direction is to extract more information from the brain response. Typical models (including those reported here) exploit low-frequency components, but useful information may also be carried by high-frequency power (Synigal et al, 2020; Forte et al, 2017; Teoh et al, 2019). Standard linear techniques (such as CCA) are not directly applicable to enhance weak sources of power, but it may be possible to use quadratic component analysis (QCA) for that purpose (de Cheveigné, 2012).…”