Evaluation of sound-event classification for robot hearing

• Main Author: Ng, Max Rui Min
• Other Authors: Jiang Xudong
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: http://hdl.handle.net/10356/77561
• Institution: Nanyang Technological University

Over the years, there have been numerous ways of carrying out the process of sound event classification. Some of these known methods consists of the use of the spectrogram, a time-frequency spectral analysis that depicts the magnitude spectrum of the signal within a 2D time-frequency plane. Though useful, there could have been further improvements that could have been made. For example, what if the phase spectrum was included as well? Would it have been able to provide more distinguishable features to the sound for effective sound classification? Therefore, in this Final Year Project, an evaluation will be carried out for a process called “Regularized 2D complex-log-Fourier transform”, which was originally proposed by Jiang Xudong and Ren Jianfeng. In the “Regularized 2D complex-log-Fourier transform”, both magnitude and phase spectrums of the signal will be utilized for the sound event classification. In addition, the Principal Component Analysis, PCA, will be used to facilitate in the removal of unnecessary features in the samples which might pose an issue during the classification process. Finally, the method chosen for the sound classification is to calculate “Mahalanobis Distance” between the test samples and the reference data of the training samples.