Swiftlet sound identification using vector quantization and gaussian mixture model

• Main Author: Siti Nurzalikha Zaini, Husni Zaini
• Format: Thesis
• Language: English
• Published: 2018
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://umpir.ump.edu.my/id/eprint/24593/1/Swiftlet%20sound%20identification%20using%20vector%20quantization%20and%20gaussian%20mixture%20model.pdf; http://umpir.ump.edu.my/id/eprint/24593/
• Institution: Universiti Malaysia Pahang

Bird sound identification has become one of the applications in audio recognition technology. Audio recognition is a great way to classify swiftlet‟s sound between baby, adult, and colony. In real life, biologists are having difficulties to identify the difference between these three types of sound except for human expert hearing experience in swiftlet farming. The identification of swiftlet sound is used to increase the production nest and quality of habitat because the main characteristic of swiftlet is its attraction toward sound. The aim of this study is to implement in swiftlet sound specifically using audio recognition to identify the types of sound. In this work, swiftlet sound feature extracted using Linear Predictive Cepstral Coefficient (LPCC), and Mel Frequency Cepstral Coefficient (MFCC) then classify the sounds using Minimum Distance Classifier (MDC), Vector Quantization (VQ) and Gaussian Mixture Model (GMM). Firstly, the features extracted using LPCC and MFCC are stored in the database. Secondly, feature extraction results in the database used for classifying the swiftlets sound using MDC, VQ with codebook size is 8, 16, 32 and 64 and GMM by 1-mixture and 2-mixture for classification. Thirdly, the best performance classification selected for an additional feature in feature extraction such as Delta and Delta-Acceleration qualifier to improve accuracy for getting a better result. Based on the result of this study, the best performance was selected based on higher accuracy identification is MFCC with GMM by 2-mixture accuracy 88.89%. At the end of the experiment, the MFCC with additional features Delta-Acceleration using classification GMM by 2-mixture with improvement 6.67% compared to original and make it up to 95.56% accuracy which is considered as good percentage result. As conclusion, the best feature extraction for swiftlet sound identification is MFCC with Delta-Acceleration features by classify the sound using GMM 2-mixture.