Related Experiment Video
Updated: Dec 24, 2025
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
922
Deep Spiking Neural Networks for Large Vocabulary Automatic Speech Recognition
Jibin Wu1, Emre Yılmaz1, Malu Zhang1
1Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore.
Frontiers in Neuroscience
|April 8, 2020
Summary
Spiking neural networks (SNNs) offer an energy-efficient alternative to artificial neural networks (ANNs) for automatic speech recognition (ASR). SNNs achieve competitive accuracy with significantly reduced computational demands, making them ideal for mobile devices.
Area of Science:
- Computer Science
- Artificial Intelligence
- Neuroscience
Background:
- Artificial neural networks (ANNs) dominate acoustic modeling in large vocabulary automatic speech recognition (ASR), but demand substantial computational resources.
- Spiking neural networks (SNNs), inspired by biological neural networks, offer potential for energy-efficient, low-power computation using spike-based processing on neuromorphic hardware.
Purpose of the Study:
- To investigate the efficacy of SNNs as an acoustic modeling technique for automatic speech recognition (ASR).
- To evaluate the performance and computational efficiency of SNNs compared to conventional ANNs in ASR tasks.
Main Methods:
- Utilized deep spiking neural networks (SNNs) for acoustic modeling in several large vocabulary speech recognition scenarios.
- Evaluated SNN performance against traditional artificial neural networks (ANNs) in terms of accuracy and computational load.
Main Results:
- SNNs demonstrated ASR accuracies comparable to their ANN counterparts.
- SNNs required only 10 algorithmic time steps and 0.68 times the total synaptic operations per audio frame compared to ANNs.
Conclusions:
- Spiking neural networks present a viable and computationally efficient alternative for acoustic modeling in automatic speech recognition.
- The integration of deep SNNs with energy-efficient neuromorphic hardware provides an attractive solution for on-device ASR applications in mobile and embedded systems.
