A biomimetic deep encoder for auditory restoration by cortical stimulation

Braincodec, a deep encoder for auditory cortical stimulation, reconstructs natural sounds with high fidelity using only a few hundred channels, outperforming cochlear implant encoders.

• Why it matters: Electrical cochlear and auditory brainstem stimulation is limited in resolution, restricting information transfer and degrading perceptual quality. Targeting auditory cortex could support hundreds of channels, but encoding algorithms that match cortical neural coding are challenging.
• What they did: Braincodec is an online deep network that jointly optimizes sound information transmission and alignment of stimulation patterns to spatial, temporal, and high-level properties of the auditory cortex neural code. It was tested in mice using a Braincodec-driven auditory cortical prosthesis for real-sound discrimination.
• The result: In mice, the prosthesis matched non-implanted animals in discriminating real sounds, supporting high-fidelity, high-throughput cortical neuroprosthetics. The approach suggests a generic framework for other sensory modalities.

AI summary of the abstract · may contain errors · not medical advice · read the paper.