Explainable Machine Learning Models and Architectures

4A Low-Power Audio Processing Using Machine Learning Module on FPGA and Applications

Suman Lata Tripathi^1,2*, Dasari Lakshmi Prasanna¹ and Mufti Mahmud²

¹VLSI Design Lab, Lovely Professional University, Punjab, India

²Department of Computer Science, Nottingham Trent University, Nottingham, UK

Abstract

Increasing demand for smart devices lead to the extensive use of machine learning techniques focused on large data processing in the form of text, audio, or image signals. The validity of such machine learning modules is based on their performance optimisation and accuracy in real-time systems. Also, low power consumption is another major challenge for any smart portable device to maintain the load on the battery and battery backup for a longer time. As such, the analysis must be performed before finalising any application-specific chip or board for such signal processing modules that can be implemented with reconfigurable architectures like field programable gate arrays (FPGA) boards. FPGA facilitates to development of real-time systems corresponding to the prototype of ML modules for different types of signals or data. This chapter gives details about the machine learning classifiers (MLC) that are frequently used for audio signal processing along with the development of real-time systems with FPGA for these modules.

Keywords: MLC, audio signal processing, FPGA, Xilinx Vivado, biomedical data

4.1 Introduction

Machine Learning Classifiers (MLC) are used in many areas; the applications ...