Portable Keyword Spotting and Sound Source Detection System Design on Mobile Robot with Mini Microphone Array

Abstract

The capability of locating sound source and detecting speech has been an essential feature in the robot auditory that enables it to and interact with human and perform sophisticated tasks. Recent development of mini microphone array has made robot audition much more accessible and offer a flexible platform to build a robot hearing system. This research proposes a keyword spotting and sound source detection system that is built on ReSpeaker Core v2.0 microphone array. We use Frequency Domain Binaural Model (FDBM) to detect the Direction of Arrival (DOA) of the sound source and estimate the distance of the sound source by taking advantage of the robot mobility to perform a triangulation method. Separate model based on Long-Short Term Memory (LSTM) network is utilized to perform the keyword spotting task. We evaluate the average accuracy of the DOA and distance and Word Error Rate (WER) of the keyword spotting activity. We also compare the performance of the system in the open field non reverberant situation with an ideal simulation.