Tether’s BrainWhisperer project marks a significant advancement in brain-computer interface (BCI) technology, achieving a variable 98.3% accuracy in decoding neural signals into text. The system uses intracranial implants combined with Tether’s QVAC AI platform to process brain activity directly on-device, optimizing privacy and efficiency.
BrainWhisperer builds on OpenAI’s Whisper automatic speech recognition model, enhanced with Low-Rank Adaptation (LoRA) fine-tuning, enabling precise phoneme decoding and progressive reductions in word error rate across multiple trial sessions.
The system was benchmarked in the 2025 Brain-to-Text Kaggle competition, ranking fourth out of 466 participants with a 1.78% word error rate, less than 0.25% behind the top performer. Tether’s multi-stage pipeline employs ensemble learning across three major datasets (Willet, Card, and Kunz) and integrates Weighted-Finite-State Transducers to optimize phoneme-to-text transcription.
This institutional-grade research demonstrates the viability of BCI-assisted text generation for clinical and accessibility applications, providing a robust framework for potential integration with healthcare or assistive technologies.
Cross-Subject and Non-Invasive BCI Applications
Beyond single-subject decoding, Tether is developing cross-subject contextual training frameworks. These systems leverage low-dimensional neural representations to translate brain activity across different individuals, reducing calibration time and resource overhead. Hierarchical Connectionist Temporal Classification models provide improved convergence and accuracy compared to standard CTC approaches, enabling more scalable, multi-user implementations.
Tether is also investigating non-invasive BCI alternatives, including surface electromyography electrodes, which detect brain and muscle signals without surgical implants. These devices aim to reduce barriers for broader adoption, increase ergonomics, and maintain transcription fidelity despite signal interference. By combining invasive and non-invasive approaches, Tether is building a versatile platform for both research institutions and end users seeking AI-driven brain-to-text capabilities.
The company emphasizes its commitment to accessibility, privacy, and institutional utility, with applications ranging from assistive communication devices to advanced neurotechnological research. BrainWhisperer represents a benchmark in AI-augmented neurotechnology, setting new standards for precision, scalability, and cross-disciplinary integration in brain-computer interfaces.