The company said training began two days after surgery and reached 71% decoding accuracy across 142 commonly used Mandarin syllables, with per-character latency below 100 milliseconds. By day seven, the system covered 418 Mandarin syllables. Decoded signals were also used for speech synthesis, a digital avatar, interaction with a large language model and control of a robotic hand.
The figures come from a company report of one case. The participant could speak, and the implant was used alongside an existing neurosurgical procedure, conditions that differ from late-stage ALS or post-stroke speech loss. The result supports feasibility but needs testing in more nonverbal patients with standardized tasks and peer-reviewed reporting.