Implementation of key phrase audio detection using binary neural network inference in a low power field programmable gate array device

  • Julius Oliver Rosal Rivera Applications Engineering Department, Lattice Semiconductor
  • Edmundo Perez Casulla Product Development and Quality Department, Lattice Semiconductor
  • Dianne Mary Abas Carandang Applications Engineering Department, Lattice Semiconductor

Abstract

The emergence of smart factories, cities, homes and mobile are driving shifts in system architectures and new applications that require intelligence at the edge. Artificial Intelligence/Machine Learning solutions are essential to meeting the requirements for this new generation AI-based edge computing applications. Convolutional Neural Network (CNN) has been widely used in image classification and object recognition. Low power, low latency and mobile applications is pushing further research on reducing the computational complexity and memory requirements. One breakthrough involves the use of binary weights and activation, which is known as Binary Neural Network (BNN). Binarized neural network replaces all multiply-accumulate operations by simple XNOR-addition operations. This architecture is implemented on a low power Field Programmable Gate Array (FPGA) for Speech Recognition application using the proprietary Mobile Development Platform with CNN accelerator cores.

Published
2019-05-09
How to Cite
[1]
J. O. Rivera, E. Casulla, and D. M. Carandang. Implementation of key phrase audio detection using binary neural network inference in a low power field programmable gate array device, Proceedings of the Samahang Pisika ng Pilipinas 37, SPP-2019-PA-20 (2019). URL: https://paperview.spp-online.org/proceedings/article/view/SPP-2019-PA-20.