A high-performance speech neuroprosthesis

Speech brain–computer interfaces (BCIs) could restore rapid communication by decoding neural activity evoked by attempted speech into text or sound. Prior demonstrations have not reached accuracies sufficient for unconstrained sentences from large vocabularies. Here, using intracortical microelectrode arrays that record spiking activity, a participant with amyotrophic lateral sclerosis who cannot speak intelligibly achieved a 9.1% word error rate (WER) on a 50-word vocabulary (2.7× fewer errors than previous state-of-the-art) and a 23.8% WER on a 125,000-word vocabulary—the first demonstration, to our knowledge, of large-vocabulary decoding. Attempted speech was decoded at 62 words per minute (WPM), 3.4× faster than the previous record and approaching natural conversation speed (~160 WPM). We also show two encouraging aspects of the neural code for speech: spatially intermixed tuning to speech articulators enabling accurate decoding from a small cortical region, and a detailed articulatory representation of phonemes that persists years after paralysis. These results chart a feasible path toward restoring rapid communication to people with paralysis who can no longer speak.

Francis R. Willett, Erin M. Kunz, Chaofei Fan, Donald T. Avansino, Guy H. Wilson, Eun Young Choi, Foram Kamdar, Matthew F. Glasser, Leigh R. Hochberg, Shaul Druckmann, Krishna V. Shenoy, Jaimie M. Henderson