Effects of acoustic periodicity and intelligibility on the neural oscillations in response to speech

Abstract

Although several studies have investigated neural oscillations in response to acoustically degraded speech, it is still a matter of debate which neural frequencies reflect speech intelligibility. Part of the problem is that effects of acoustics and intelligibility have so far not been considered independently. In the current electroencephalography (EEG) study the amount of acoustic periodicity (i.e. the amount of time the stimulus sentences were voiced) was manipulated, while using the listeners' spoken responses to control for differences in intelligibility. Firstly, the total EEG power changes in response to completely aperiodic (noise-vocoded) speech and speech with a natural mix of periodicity and aperiodicity were almost identical, while an increase in theta power (5-6.3Hz) and a trend for less beta power (11-18Hz) were observed in response to completely periodic speech. These two effects are taken to indicate an information processing conflict caused by the unnatural acoustic properties of the stimuli, and that the subjects may have internally rehearsed the sentences as a result of this. Secondly, we separately investigated effects of intelligibility by sorting the trials in the periodic condition according to the listeners' spoken responses. The comparison of intelligible and largely unintelligible trials revealed that the total EEG power in the delta band (1.7-2.7Hz) was markedly increased during the second half of the intelligible trials, which suggests that delta oscillations are an indicator of successful speech understanding.

Predicting the effects of periodicity on the intelligibility of masked speech: An evaluation of different modelling approaches and their limitations

“…This alteration is based on the theoretical consideration that speech carries no relevant linguistic information at these modulation frequencies (Arnal et al, 2015;Joris et al, 2004), neurophysiological evidence that there is little activity in response to these modulation rates in human auditory cortex (Giraud et al, 2000), and the empirical observation that the speech modulation spectrum indeed shows a dip in this region (e.g., Fig. 7 in Steinmetzger and Rosen, 2017). Rather than discarding them a priori, an algorithm was developed that identifies and excludes any modulation filters from further processing for which the broadband modulation power of the unprocessed speech signal falls below a specified relative threshold.…”

Electroencephalographic correlates of word and non-word listening in children with specific language impairment: An observational study

“…[ 76 ] The third limitation also refers to the EEG design study regarding the use of spectral power only. A majority of studies [ 3 , 49 , 84 , 85 ] that explored the neural basis of auditory processing have used even-related potential measures and measures with MMN response.…”

Formant‐invariant voice and pitch representations are pre‐attentively formed from constantly varying speech and non‐speech stimuli

“…Rotated speech stimuli were created by rotating the spectrum of speech stimuli using a spectral rotation function in MATLAB (MATLAB, 2020) with a cut‐off frequency of 4000 Hz (available at https://www.phon.ucl.ac.uk/downloads/matlab/Blesser.zip); the same function and other similar implementations of the spectral rotation algorithm were used in several studies to produce non‐speech control stimuli in the attempt to contrast acoustic and speech‐specific perceptual processes (Azadpour & Balaban, 2008; Marklund et al, 2020; Scott, 2000; Steinmetzger & Rosen, 2017). The result of this procedure is a sound with a mirrored spectrogram along a mirroring frequency (i.e., 2000 Hz corresponding to half of the cut‐off frequency) with respect to the input sound.…”