UCSF brain-computer interface decodes speech and body gestures simultaneously from a single cortical implant
A Nature Neuroscience paper published on 14 September 2026 by Samantha C. Brosler, Jessie R. Liu, Alexander B. Silva and senior author Edward F. Chang at UCSF reports the first single-implant, single-session, real-time simultaneous decoding of speech and upper-body gestures in people with paralysis. Two of three BRAVO-trial participants (Bravo-1r, brainstem stroke, seven years enrolled; Bravo-6, ALS, implanted March 2025) controlled a personalized virtual avatar with parallel speech and gesture decoders reading from a 253-channel high-density ECoG surface array manufactured by PMT Corporation. Bravo-1r reached median 100 percent accuracy on both decoders in the real-time conversation paradigm; Bravo-6 reached 85 percent gesture and 75 percent speech. The design finding is that multimodal neural signals are not the linear sum of isolated speech and gesture signals: decoders trained on simultaneous data outperform decoders trained on isolated data, and cross-training drops opposite-modality false-positive rates from 76 percent (Bravo-1r) and 31 percent (Bravo-6) to zero. Chang has stated a fully implantable wireless successor is the next step. Hardware footprint is PMT surface ECoG plus a Blackrock Neurotech digital headstage per prior Chang-lab methods, with no overlap with Neuralink Threads, Paradromics Connexus, or Synchron Stentrode. This is the fourth UCSF Chang-lab BRAVO paper in the peer-reviewed communication-BCI sequence: Moses NEJM 2021, Metzger Nature 2023, Littlejohn Nature Neuroscience 2025, Brosler Nature Neuroscience 2026.