Tether has introduced BrainWhisperer, a brain-computer interface system designed to translate neural signals into text with up to 98.3% decoding accuracy. The technology combines intracranial implants with Tether’s QVAC AI platform to process brain activity locally on devices, improving privacy and efficiency.
BrainWhisperer builds on the architecture of Whisper, enhanced through Low-Rank Adaptation fine-tuning to enable precise phoneme decoding. The system achieved a 1.78% word error rate in the 2025 Brain-to-Text Kaggle competition, placing fourth among 466 participants. Its pipeline integrates ensemble learning across multiple neural datasets and uses Weighted-Finite-State Transducers to optimize phoneme-to-text transcription.
Beyond single-user experiments, Tether is developing cross-subject models capable of interpreting neural signals from different individuals with reduced calibration requirements. The company is also exploring non-invasive alternatives such as surface electromyography sensors that capture neural and muscle signals without surgical implants.
This also aligns with Tether’s broader push into AI-powered health technologies, following its recent investment in Eight Sleep at a $1.5 billion valuation to expand personalized health intelligence and wellness analytics powered by artificial intelligence.
