Keyword spotting in audio data for edge devices using Machine learning algorithms on an FPGA platform.

Number of students: 2
Guide: K R Viveka

Keyword detection in audio streams is an important challenge with applications ranging from automatic speech recognition to building better human machine interface. The availability of multiple microphones in edge devices can be leveraged to improve both noise performance and identify directionality of audio data. This project will target the implementation of a machine learning algorithm on an FPGA platform to analyse audio information. The algorithm will need customisation and optimisation to minimise its hardware footprint while meeting accuracy targets set by the application. Further optimisations include making the system robust against deliberate distractions, improving data security, analysing trade offs between system overhead and keyword range, and improving noise resilience, which may be explored based on available bandwidth.