Scientists from Stanford have achieved a very high language decoding rate, setting a new record in the brain-computer interface (Brain-Computer Interface, BCI).

The system, which recorded peak activity from intracortical microelectrode arrays, enabled a study participant who could no longer speak clearly due to amyotrophic lateral sclerosis to use a vocabulary of 50 words and a vocabulary of 125,000 words with an error rate of 9.1%. but with 23.8% error rates. At the same time, the record is that the speech decoding speed reaches 62 words per minute, which is not only fast, but 3.4 times faster than the previous record for any type of BCI. For comparison, the average rate of natural speech is about 160 words per minute.

The scientists state that the improved language model will reduce the error rate to 17.4% for a 125,000-word dictionary, and some methods can reduce this rate to 11.8%. The article published by the scientists is now a preprint.