referenceArchiveUUID: "72ed291f-4dc7-4e49-b3ce-55abdbbb46be"
fullTitle: "A high-performance speech neuroprosthesis"
authors: ["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"]
https://doi.org/10.1038/s41586-023-06377-x
Intracortical BCI using microelectrode array, attempted and internal speech. Achieved 9.1% WER on a 50 word vocabulary and 23.8% WER on 125k word vocabulary (“first successful demonstration of large-vocabulary decoding”). Decoded speech at 62 WPM. All these figures are integer-factor improvements over previous SotA.
Pretty simple model, at a high level: threshold crossings are decoded by RNN into phonemes; then onward into Kaldi as usual for ASR.
Same first and senior authors as the very impressive earlier:
High-performance brain-to-text communication via handwriting. Francis R. Willett, Donald T. Avansino, Leigh R. Hochberg, Jaimie M. Henderson & Krishna V. Shenoy. https://www.nature.com/articles/s41586-021-03506-2)