Intelligent audio surveillance using audio acquisition, blind source separation, and audio event detection

• Main Author: Dadula, Cristina P.
• Format: text
• Language: English
• Published: Animo Repository 2017
• Subjects: Algorithms; Sound analyzers; Sound > Equipment and supplies; Audio equipment test recordings; Engineering
• Online Access: https://animorepository.dlsu.edu.ph/etd_doctoral/515
• Institution: De La Salle University

This study presents the design and implementation of an audio surveillance system using audio acquisition, blind source separation, and audio event detection. The audio acquisition from multiple USB microphones was designed for future development of an embedded computer system of some real time audio processing and analysis. It was designed using multiple threads and was implemented in Java. Thread is a sequential program that has its own thread of control and can be executed concurrently with other threads. The algorithm successfully captured audio data from multiple USB microphones. The design and implementation of blind source separation was based on independent component analysis (ICA). Independent component analysis, such as kurtosis and mutual information, were used to measure in separating the independent sources. The estimated sound sources were obtained based on the maximization of kurtosis and minimization of mutual information. The optimization process was done by using genetic algorithm and was implemented in Java programming language. The simulation was successful in separating sources up to 4 mixed signals. However, the algorithm did not work well in real recorded signals because the coefficients obtained were not enough to represent the demixing matrix. The audio event detection algorithms were implemented using mel frequency cepstral coefficients as feature vector of the audio signals. Three different classifiers were designed: adaptive neuro fuzzy inference system, neural network, and fuzzy inference system.