Deep-learning algorithms for speech rehabilitation with implantable brain-computer interfaces

Société : INSERM U1216 - Institut des Neurosciences de Grenoble

The Grenoble Institute for Neuroscience is a joint research unit from the Institut National pour la Santé et la Recherche Médicale (INSERM) and the University Grenoble Alpes. The laboratory collaborates closely with the CEA-Clinatec institute to implement a chronic speech BCI clinical trials for LIS participants.

Description du poste

Project: Approximately 18.5 million individuals have a speech, voice, or language disorder worldwide, and about 300,000 in France. Among them, several millions have speech disability secondary to neurodegenerative diseases (such as Amyotrophic Lateral Sclerosis – ALS) or brainstem strokes damaging motor pathways dedicated to speech articulation and voicing. These impairments may result in a locked in syndrome (LIS), where people are unable to speak while their cognitive abilities remain intact. In this context, brain-computer interface (BCI) systems aim to provide communication solutions for these people by recording their brain signals and “translating” them into artificial speech produced by a synthesizer [1], [2], [3], [4], [5], [6], [7], [8]. Our current research is focused on the development of a fully implantable speech BCI that reconstructs speech from intracranial electrocorticographic (ECoG) signals in real time using deep learning [9], [10], [11]. The goal of the internship will be to develop deep neural network algorithms to decode speech from ECoG signals and make these algorithms compliant with a dedicated real-time software environment developed in the laboratory.

References:
[1] F. Bocquelet, T. Hueber, L. Girin, S. Chabardes, et B. Yvert, « Key considerations in designing a speech brain-computer interface », Journal of Physiology Paris, vol. 110, no 4, p. 392‑401, 2016, doi: 10.1016/j.jphysparis.2017.07.002.
[2] F. Bocquelet, T. Hueber, L. Girin, C. Savariaux, et B. Yvert, « Real-Time Control of an Articulatory-Based Speech Synthesizer for Brain Computer Interfaces », PLOS Computational Biology, vol. 12, no 11, p. e1005119, nov. 2016.
[3] P. Roussel et al., « Observation and assessment of acoustic contamination of electrophysiological brain signals during speech production and sound perception », Journal of Neural Engineering, vol. 17, p. 056028, 2020, doi: 10.1088/1741-2552/abb25e.
[4] D. A. Moses et al., « Neuroprosthesis for Decoding Speech in a Paralyzed Person with Anarthria », New England Journal of Medicine, vol. 385, no 3, p. 217‑227, 2021, doi: 10.1056/nejmoa2027540.
[5] S. L. Metzger et al., « A high-performance neuroprosthesis for speech decoding and avatar control », Nature, vol. 620, no 7976, p. 1037‑1046, 2023, doi: 10.1038/s41586-023-06443-4.
[6] F. Willett et al., « A high-performance speech neuroprosthesis. », Nature, vol. 620, p. 1031‑36, 2023, doi: 10.1038/s41586-023-06377-x.
[7] G. Le Godais et al., « Overt speech decoding from cortical activity: a comparison of different linear methods », Frontiers in Human Neuroscience, vol. 17, no June, p. 1124065, 2023, doi: 10.3389/fnhum.2023.1124065.
[8] N. S. Card et al., « An Accurate and Rapidly Calibrating Speech Neuroprosthesis », N Engl J Med, vol. 391, no 7, p. 609‑618, août 2024, doi: 10.1056/NEJMoa2314132.
[9] X. Ran, W. Chen, B. Yvert, et S. Zhang, « A hybrid autoencoder framework of dimensionality reduction for brain-computer interface decoding », Computers in Biology and Medicine, vol. 148, p. 105871, 2022, doi: https://doi.org/10.1016/j.compbiomed.2022.105871.
[10] Z. Wan et al., « An Interpretable and Generalizable Speech Detector Based on a CNN-LSTM Framework », in ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), avr. 2024, p. 13231‑13235. doi: 10.1109/ICASSP48485.2024.10445835.
[11] M. Benticha et al., « A Vision Transformer Architecture For Overt Speech Decoding From ECoG Data », présenté à 46th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), Orlando, juill. 2024.

Profil recherché

Profile: Candidates should be highly motivated with a solid training in computer science, machine learning, artificial intelligence, and signal processing. Proficient programming skills in Matlab/Python are required. A good autonomy and ability for interdisciplinary team work and fluent proficiency in English writing and speaking are mandatory while no specific knowledge of French is required. This internship may open to a PhD thesis project.