edwardzcl / RSNN

Residual Spiking Neural Network for Speech Keyword Spotting

Residual Spiking Neural Network on a Programmable Neuromorphic Hardware for Speech Keyword Spotting

---

This code can be used as the supplemental material for the paper: "Residual Spiking Neural Network on a Programmable Neuromorphic Hardware for Speech Keyword Spotting". (Published on IEEE ICSICT, October, 2022).

---

Citation:

C. Zou, X. Cui, S. Feng, G. Chen, X. Wang and Y. Wang, "Residual Spiking Neural Network on a Programmable Neuromorphic Hardware for Speech Keyword Spotting," 2022 IEEE 16th International Conference on Solid-State and Integrated CIrcuit Technology (ICSICT), 2022, pp. 1-3, doi: xxx.

Features:

• This supplemental material gives a reproduction function of training and testing experiments of the improved residual RNN (R-SNN) in our paper. Totally, three kinds of optional residual architectures are compared.

File overview:

• README.md - this readme file.
  
• data - the speech dataset folder.
  
• figs - the visualized figure folder.
  
• tensorlayer - the provided binary/ternary package, named tensorlayer.
  
• tools - some available scripts.
  
• k0B2_asr_shortcut_group4_noplace_A.py - the training script for the traditional residual architecture i.e. A in our paper.
  
• k0B2_asr_shortcut_group4_noplace_B.py - the training script for the traditional residual architecture i.e. B in our paper.
  
• k0B2_asr_shortcut_group4_noplace_C.py - the training script for the improve residual architecture in our paper.
  
• Spiking_asr_shortcut.py - the residual SNN inference script.
  
• spiking_ulils.py - the tool script for various spiking operators.
  

Requirements:

1. Python-3.6, librosa-0.4
   
2. Tensorflow 1.2 for cpu or gpu
   
3. CPU or GPU server

Usage:

• Please note you have installed the package Tensorflow=1.2.x, then directly run with:

$ python k0B2_asr_shortcut_group4_noplace_C.py.py --k 0 --B 2 --learning_rate 0.01 --resume False --mode 'training'

for the improve residual architecture training, or

$ python Spiking_asr_shortcut.py --k 0 --B 2 --noise_ratio 0 --learning_rate 0.01 --resume True --mode 'testing'

for the improve residual architecture testing.

Results

• Please refer to our paper for more information.

More question:

• There might be a little difference of results for multiple training repetitions, because of the randomization.
• Please feel free to reach out here or email: [e-mail] if you have any questions or difficulties. I'm happy to help guide you.

Reference

• Ternary Weight Networks
• Tensorlayer
• Scatter-and-gather scheme