A speech envelope landmark for syllable encoding in human superior temporal gyrus

Abstract

Listeners use the slow amplitude modulations of speech, known as the envelope, to segment continuous speech into syllables. However, the underlying neural computations are heavily debated. We used high-density intracranial cortical recordings while participants listened to natural and synthesized control speech stimuli to determine how the envelope is represented in the human superior temporal gyrus (STG), a critical auditory brain area for speech processing.We found that the STG does not encode the instantaneous, moment-by-moment amplitude envelope of speech. Rather, a zone of the middle STG detects discrete acoustic onset edges, defined by local maxima in the rate-of-change of the envelope. Acoustic analysis demonstrated that acoustic onset edges reliably cue the information-rich transition between the consonantonset and vowel-nucleus of syllables. Furthermore, the steepness of the acoustic edge cued whether a syllable was stressed. Synthesized amplitude-modulated tone stimuli showed that steeper edges elicited monotonically greater cortical responses, confirming the encoding of relative but not absolute amplitude. Overall, encoding of the timing and magnitude of acoustic onset edges in STG underlies our perception of the syllabic rhythm of speech.

Auditory features modelling reveals sound envelope representation in striate cortex

“…Conversely, it was shown that portions of STG encode the amplitude variations of speech as well as amplitudemodulated tone stimuli independent from other spectro-temporal features (Oganian and Chang, 2019). Taken together, these results reveal that in the superior temporal cortex, regions coding speech-specific spectrotemporal features and non-speech-specific envelope variations coexist.…”

“…Do High and Low frequency ranges share a common spatial organization in the left hemisphere across linguistic materials? Previous results showed that the sound envelope is represented in speech-related cortical areas within the left hemisphere (Giraud, 2000;Oganian and Chang, 2019). An antero-posterior gradient from low to high modulation frequencies emerged (DeWitt and Rauschecker, 2012;Hullett et al, 2016).…”

“…Sound envelope is known to be represented in speech-related cortical areas within the left hemisphere, with antero-posterior gradient from low to high modulation frequencies (Giraud, 2000;Oganian and Chang, 2019;DeWitt and Rauschecker, 2012;Hullett et al, 2016). To assess the degree of spatial overlap of the goodness of fit for the High and Low frequency ranges, we measured the correlation of the RMSE between them, for each speech-related sound category (words, pseudowords, artificial), across all voxels in the occipital and temporal ROIs ( Figure 5A).…”

Phase alignment of low-frequency neural activity to the amplitude envelope of speech reflects evoked responses to acoustic edges, not oscillatory entrainment

“…We build upon recent intracranial electrocorticography (ECoG) recordings that showed that local 8 neural populations on the superior temporal gyrus (STG) selectively respond to acoustic edges in the 9 speech envelope and that envelope tracking by broadband high gamma amplitude was explained by 10 transient power increases following acoustic edges (Oganian & Chang, 2019). Results from this study 11 suggested that onset edges may be the primary acoustic feature driving speech envelope encoding.…”

Phoneme-level processing in low-frequency cortical responses to speech explained by acoustic features

“…Another possibility is that the performance boost of the articulatory features is instead attributable to their correlation with a low-dimensional and less rich acoustic feature. It has repeatedly been observed that neuronal responses from bilateral superior temporal regions are especially sensitive to acoustic edges (Prendergast et al, 2010;Hertrich et al, 2012;Gross et al, 2013;Doelling et al, 2014;Oganian & Chang, 2018). Corresponding feature spaces extracting these onsets from envelope representations by means of half-wave rectifying the temporal gradient have previously been used by several labs (Hertrich et al, 2012;Hambrook & Tata, 2014;Fiedler et al, 2017).…”