Individual theta-band cortical entrainment to speech in quiet predicts word-in-noise comprehension

Abstract

Speech elicits brain activity time-locked to its amplitude envelope. The resulting speech-brain synchrony (SBS) is thought to be crucial to speech parsing and comprehension. It has been shown that higher speech-brain coherence is associated with increased speech intelligibility. However, studies depending on the experimental manipulation of speech stimuli do not allow conclusion about the causality of the observed tracking. Here, we investigate whether individual differences in the intrinsic propensity to track the speech envelope when listening to speech-in-quiet is predictive of individual differences in speech-recognition-in-noise, in an independent task. We evaluated the cerebral tracking of speech in source-localized magnetoencephalography, at timescales corresponding to the phrases, words, syllables and phonemes. We found that individual differences in syllabic tracking in right superior temporal gyrus and in left middle temporal gyrus (MTG) were positively associated with recognition accuracy in an independent words-in-noise task. Furthermore, directed connectivity analysis showed that this relationship is partially mediated by top-down connectivity from premotor cortex-associated with speech processing and active sensing in the auditory domain-to left MTG. Thus, the extent of SBS-even during clear speech-reflects an active mechanism of the speech processing system that may confer resilience to noise.

Keywords: cerebro-acoustic coherence; functional connectivity; granger causality; speech comprehension; speech-in-noise.

Fig. 1

Average modulation spectrum and spectral ranges of stimulus material (Aesop’s fables) used in this study. Before averaging across stories, the normalized modulation spectrum for each short story was calculated as in Ding et al. (2017a, 2017b), bands were derived as in Keitel et al. (2018), spectral ranges are indicated by the colored fill-ins. NB: There is overlap between word and syllable range (as indicated by the blueish color shade).

Fig. 2

Descriptive overview of demographics of population and distribution of the behavioral-cognitive variables used in this study. This matrix shows age, gender, and the participants’ scores for WiN recognition. The diagonal shows the distribution of individual values for the population characteristics and the transformed WiN measures (denoted WiN’, see Materials and Methods for details of transformation). Original WiN scores are shown in the last column. Correlations (linear Pearson’s and point-biserial correlation) and scatter plots are derived from the transformed version of WiN. Linear fits are shown in red, with confidence intervals (±95%) shown in gray. Units of the original measures: WiN—in dB SNR. Transformed data (WiN’) are Gaussianized by rank-based inverse normalization and also normalized to zero-mean and unit variance (see Materials and Methods). No significant correlations were observed.

Fig. 3

Visualization of the edges (connections) used for granger causality and subsequent analyses. Top: 1 network consists of 5 parcels for each hemisphere, comprising frontal areas (superior, inferior frontal gyrus and PRG regions) involved in language processing, connected to down-stream regions of interest that are defined by significant SBS (termed SBS-ROIs), centered in left MTG and right STG. Bottom: a second network includes 2 parcels centered in left and right primary auditory cortex and their connections to the two SBS-ROIs.

Fig. 4

Significant SBS (P < 0.05), as measured by GCMI, observed at 3 different frequency bands. At phrasal rate, left temporal areas show stronger SBS than corresponding right areas and there are also extensive subcortical/thalamic foci, whereas at faster rates, i.e. word and syllable rate, the focus of SBS shifts to right temporal areas including primary auditory cortex. Please note that mutual information is not frequency independent, and tends to become larger with decreasing frequency, as indicated by differences in scale.

Fig. 5

Visualizing the relationship of SBS vs WiN, showing the brain areas with significant relationship between speech tracking and WiN comprehension, across all rates that showed a significant relationship (and thresholded at P < 0.05). Both at phrasal, word and syllable rate, there was a negative correlation between speech tracking and WiN comprehension. At phrasal rate (A), bilateral subthalamic areas, left ventral anterior cingulate and right STG were observed. At word rate (B), medial orbitofrontal areas were showing significant SBS, whereas at syllable rate (C), SBS in a right cluster in STG was significant. Scatter plots on the right illustrate the negative correlation between SBS and WiN at the different rates, all indicating better recognition performance (= lower WiN score) with higher speech tracking in quiet. At each rate, the chosen cluster is indicated by an orange-lined box.

Fig. 6

CCA of the relationship between TD/bottom-up connectivity and SBS and WiN yields 2 canonical modes with significant relationships. (A–C) For TD connectivity from frontal to SBS-ROIs (A), CCA yielded 1 significant mode in a left-hemisphere parcel. Its spectrally resolved connectivity profile is composed of beta and lower gamma activity (B) upmodulating SBS and down-modulating WiN scores (= positive impact on WiN performance), whereas lower-frequency (alpha and below) activity shows inverse relationship to SBS and WiN (= negative impact), see c. (D–F) For connectivity between primary auditory cortex and SBS-ROIs (left MTG and right STG), we observed 1 canonical mode, from left primary auditory cortex to left MTG SBS-ROI (D), with a spectral profile showing mostly activity in the gamma range (E). This mode is significantly up-modulating SBS, however was not significantly linked to WiN (F). Abbreviation: A1, primary auditory cortex. Asterisks indicate significance of post hoc single-variable correlations (^* = P < 0.05, ^** = P < 0.01, ^*** = P < 0.001). For connectivity (B, E), thick lines indicate correlation coefficients with the neighboring thin lines surrounding it indicating the 95% confidence interval. Analogously, the bar plots for WiN and SBS (C, F) are complemented by 95% confidence intervals as well in the form of error bars.