Spiking network adapts computation to improve speech command recognition
SpikCTU: A Compact Spiking Network with Adaptive Recurrent Computation for Speech Command Recognition
Sound
Summary
Recognizing spoken commands using computers can be energy-intensive and sometimes inefficient. The authors developed SpikCTU, a small and efficient neural network that mimics brain spikes and adapts how much computing it does based on how hard a speech command is to recognize. This approach not only saves energy but also improves accuracy by refining its guesses step-by-step. The method also lets the system stop early if it is confident enough, saving even more computing power.
What this means in practice
- •For mobile app developers: Implement low-power speech command recognition that dynamically adjusts computing effort to save battery while maintaining accuracy.
- •For voice assistant designers: Design compact speech recognition modules that use fewer parameters and can stop processing early when confident, reducing latency and compute cost.
Authors
Li Jin, Jiang Wenbin, Chen Tao, Wen Fei
Abstract
Spiking Neural Networks offer event-driven, energy-efficient computation for speech command recognition, yet most existing methods follow fixed computation paths, allocating redundant computation to easy samples while providing limited insight into how evidence evolves during inference. To address these challenges, we introduce SpikCTU, a compact spiking network with adaptive recurrent computation for speech command recognition. Specifically, a hybrid spiking embedding module preserves continuous input information alongside spike features to construct a fixed temporal memory. A shared Continuous Thinking Unit (CTU) repeatedly reads this memory using keys derived from spikes and continuous values, and updates a binary recurrent state with lightweight stage adapters. Furthermore, combining a fixed base prediction with accumulated corrections at each step enables learned aggregation over all steps or confidence-based early exit using only the predictions available so far. Experiments on three speech command recognition benchmarks show that SpikCTU achieves competitive accuracy with only 0.16--0.17M parameters, and the early exit strategy can substantially further reduce computational cost without compromising accuracy. In addition, trajectory and attention analyses reveal that recurrent steps enable error correction and progressively refine predictions. Code is available at https://github.com/Rivflyyy/spikctu.