AI and the World's Hard Problems · Theoretical computer science · arXiv · 2026-10-22
Bio-Inspired Auditory Processing: Coupling Parallel Resonate-and-Fire Dynamics with Ternary Recurrent Networks for Speech Recognition
Spiking neural networks promise energy-efficient inference but struggle with continuous, context-dependent tasks like automatic speech recognition, partly because LIF neurons lose information across pauses and phonetic boundaries. The Spiking Ternary GRU (stGRU) lets a ternary-weight gated recurrent unit perform temporal filtering while downstream resonate-and-fire neurons generate the binary spike stream, avoiding dense floating-point matrix MACs. Trained end-to-end on LibriSpeech-960, Small, Medium, and Large stGRUs reach 8.09%, 7.04%, and 6.44±0.17% WER on test-clean without an external language model, the first matrix-MAC-free spiking networks without a learned ANN sequence decoder to achieve single-digit WER on LibriSpeech.