Tether Releases Open-Source BrainWhisperer Neural Engine, Achieves 8.7% Word Error Rate

According to mpost.io, Tether today released BrainWhisperer, an open-source brain-to-text engine capable of decoding neural signals into text entirely on local devices. CEO Paolo Ardoino announced the system is fully integrated into QVAC, the company's on-device AI stack, and available to developers as a proof-of-concept capability.

The SDK version achieved an 8.7% word error rate in validation tests using real neural recordings, crossing the ten percent threshold researchers benchmark for real-world usability. The engine runs in under two gigabytes of memory with approximately fifty milliseconds latency. A more complex variant of the same architecture ranked fourth among 466 competing teams in an international brain-to-text challenge. Tether emphasizes that local execution prevents neural data transmission to external servers, addressing privacy concerns. The release is described as foundational rather than consumer-ready, as the model was trained on data from only four participants.