Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions

Neural geometry of speech preparation in ventral premotor cortex

Poster Session C, Friday, October 25, 4:30 - 6:00 pm, Great Hall 3 and 4

Benyamin Meschede-Krasa¹, Erin Kunz^2,5, Foram Kamdar³, Nicholas S. Card¹⁰, Maitreyee Wairagkar¹⁰, Carrina Iacobacci¹⁰, Samuel R. Nason-Tomaszewski¹³, Payton H. Bechefsky¹³, Leigh R. Hochberg^7,8,9, David M. Brandman¹⁰, Sergey Stavisky¹⁰, Nicholas AuYong¹², Chethan Pandarinath^11,12,14,, Jaimie M. Henderson^3,5, Francis R. Willett^2,3,5, Shaul D. Druckmann⁴; ¹Neurosciences Graduate Program, Stanford University, Stanford, CA, USA, ²Department of Electrical Engineering, Stanford University, Stanford, CA, USA, ³Department of Neurosurgery, Stanford University, Stanford, CA, USA, ⁴Department of Neurobiology, Stanford University, Stanford, CA, USA, ⁵Wu Tsai Neurosciences Institute, Stanford University, Stanford, CA, USA, ⁶Howard Hughes Medical Institute, ⁷School of Engineering and Carney Institute for Brain Sciences, Brown University, Providence, RI, USA, ⁸VA RR&D Center for Neurorestoration and Neurotechnology, VA Providence Healthcare, Providence, RI, ⁹Center for Neurotechnology and Neurorecovery, Department of Neurology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, ¹⁰Department of Neurological Surgery, University of California Davis, Davis, CA, USA, ¹¹Department of Cell Biology, Emory University, Atlanta, Georgia, USA;, ¹²Department of Neurosurgery Emory University, Atlanta, Georgia, USA;, ¹³Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, Georgia, USA;

Recent speech decoding brain-computer interfaces have shown impressive performance by predicting sequences of phonemes from ventral premotor cortex (area 6v)[1, 2]. However, the neural representations underlying the preparation and production of speech in these areas remains poorly understood. The GODIVA model [3] predicts that ventral premotor cortex, including parts of area 6v, encodes a speech sound map which serves as a high level representation of sequences of speech sounds. Here we provide empirical evidence that 6v encodes whole sequences of phonemes before speech onset using neural representations that are shared across phoneme positions and task epochs (cue perception, preparation, and execution), consistent with 6v playing a higher level role in speech production that goes beyond feedback control of the speech articulators. Microelectrode arrays were placed in inferior 6v in three Braingate2 participants. In an instructed delay task, participants were cued with audio recordings of nonsense words composed of balanced sequences of three consonants separated by a repeated vowel. An example sequence of K-N-T was cued with the ARPABET phoneme sequence K-AH-N-AH-T. After a delay period, participants attempted to reproduce the cue. LDA models were fit to predict consonant phonemes individually for each position from delay period neural activity. All 3 phoneme positions were reliably decoded (classifier AUROCs consonant 1: 0.98, consonant 2: 0.89, consonant 3: 0.97) indicating that 6v can simultaneously encode phonemes in all positions. Models were then used to predict phonemes outside of the fitted position to assess similarity in phoneme encoding across position. Decoders fit between future consonants (positions 2 and 3) could generalize across context. Projecting held out trials into LDA coding dimensions revealed a shared neural geometry for future phoneme encoding. Next we fit decoders to each of 3 contexts (cue listening, delay, and execution) and found that decoders could generalize across context, revealing a representation of speech sounds not strictly limited to motoric preparation and execution of speech. These results demonstrate that in the speech planning hierarchy, area 6v encodes a generalized preparatory representation of sequences of speech sounds. [1] Willett, F.R., Kunz, E.M., Fan, C. et al. A high-performance speech neuroprosthesis. Nature (2023). [2] Card, N.S. et al. An accurate and rapidly calibrating neuroprosthesis. Medrxiv (2023). [3] Bohland, JW., Bullock, D., & Guenther, F. H. (2010). Neural representations and mechanisms for the performance of simple speech sequences. Journal of cognitive neuroscience.

Topic Areas: Speech Motor Control,