The brief

Analyst view

• ECoG array decodes speech signals; synthesises speech via avatar.
• Achieved 78 wpm text, 25% error rate in trials with <2 weeks training.
• Neurotech Leaders 2025 award; a16z $50M Series A (Jan 2025).

Technical approach

As reported

Cortical surface recording

ECoG electrode arrays

Surface arrays on speech cortex; better SNR than deep electrodes, less invasive.

Speech decoding

Real-time neural-to-speech algorithms

Machine learning models decode speech-related neural activity to generate text, audio speech via synthesis and facial avatar animation in real time.

Output modalities

Text, speech audio and avatar

System provides three complementary outputs: typed text (78 wpm), synthesised speech audio and avatar facial animation for natural communication.