Towards audio-assist cognitive computing : algorithms and applications

Abstract

Meaningful information hidden in the acoustic signals can be utilized by cognitive computing algorithms. The algorithms use them to improve the quality of services and applications. Inspired by this idea, we develop and optimize a series of applications based on cognitive computing algorithms. Two cognitive computing algorithms are developed: Audio Tag and Audio Fingerprint algorithms. The implementation and experiment results of the algorithms suggest that the information hidden in acoustic signals, either manually implanted or innate, can be utilized by proper techniques. The experiment results demonstrate that the audio tag and audio fingerprint algorithm have high accuracy and low time cost. The audio tag algorithm achieves 100\% accuracy (recognition under 5 seconds), with loud noises existing in specific experiment environments. The audio fingerprint algorithm achieves over 95\% accuracy(recognition under 5 seconds), with proper parameter settings. Based on the two core algorithms, two android applications are developed: Hey!Shake and Parking Loud application. They utilize these algorithms in the TV watching and parking lot access control scenarios and provide services with better quality, less hardware cost, and more convenience for users. The results of this research project confirm the possibility that we can improve the quality of multimedia services by digging into the often-overlooked acoustic information.