Slide Slam

< Slide Slam Sessions

Slide Slam I2

Neuronal long-range temporal correlations are correlated with scaling laws of speech envelopes

Slide Slam Session I, Wednesday, October 6, 2021, 5:30 - 7:30 pm PDT Log In to set Timezone

Chun-Hsien Hsu1, Ting-Yu Yu2, Zi-Jun Su1; 1National Central University, 2University of Taipei

Magneto- and electroencephalographic (M/EEG) studies have demonstrated the cortical entrainment to syllabic rhythm of speech and linguistic structures of phrases. Previous studies on cortical entrainment to speech were mainly based on fitting a response function describing the linear mapping between natural speech and M/EEG signals. We aimed to extend the understanding of cortical entrainment by using the detrended fluctuation analysis (DFA), which have been widely applied to the detection of long-range temporal correlations (LRTC) in time series, such as M/EEG signals, the amplitude envelopes of M/EEG and response times. In the present study, twelve adults who are native speakers of Mandarin Chinese participated in a speech comprehension task, in which they would hear normal speeches or noise-vocoded speeches which were vocoded with 16 channels or 4 channels. Speech materials were based on a large-scale Chinese corpus (Sinica COSPRO corpus). The results demonstrated that the LRTC scaling exponents of MEG signals were significantly correlated with those of speech envelopes. This finding supports the prediction that the power-law scaling of brain activity and speech envelopes would be correlated with each other. Furthermore, we investigated whether the intelligibility of speech sound (normal, 16 channels, and 4 channels) can be predicted from spatial patterns of LRTC scaling exponents of MEG signals. Therefore, multivariable pattern analyses (MVPA) were conducted to predict stimuli types. The results demonstrated that scalp distributions of LRTC scaling exponents of MEG signals can decode the intelligibility of speech stimuli.

< Slide Slam Sessions

SNL Account Login

Forgot Password?
Create an Account