Abstract
Singapore is diverse in terms of culture and ethnicity. As a result, the local music in Singapore is an amalgam inspired from all the main ethnic groups present here, namely the Chinese, Malay and Indian (which consist of Hindi and Tamil groups). The popular music heritage leads to an unique Singaporean identity and this diverse culture impacts many factors in the country. Musicology is the scholarly analysis and research based study of music. Information from music is extracted and studied to verify and understand different hypotheses. One example is we can analyse music and understand the current culture and community. In this thesis, we propose to create a classification system to classify songs based on four ethnic groups and they are Chinese, Malay, Hindi and Tamil groups. We will use machine learning algorithms to train models using two types of approaches. First of all, we will use Melspectrogram images to train a Convolutional Neural Network model and secondly we will use feature inputs to train other traditional models. Feature inputs are independent variables that include both low-level signal properties and high-level perceptual properties. Low-level features include properties like the spectral features and Mel-frequency ceptral coefficients and the high-level features include loudness and pitch properties. We will compare all the models in terms of accuracy and other metrices like the F1score, Precision, Recall and AUC/ROC scores. To accomplish our goal, first of all we will collect songs from the four different groups and extract features and tags from the songs using popular open-source libraries. We will use feature selection methods to filter which will allow us to only use features that are important to train the models. We will also analyse and verify that not all features are necessary to obtain a good performing model and if we use feature selection methods to only use the features of importance, our model’s performance will improve significantly. Finally, we will use the best performing model to analyse the National Day Parade songs of Singapore to discover how much the songs are influenced by the chosen ethnic groups. As a conclusion, we talk about how we can improve the model and how this thesis can be a foundation to aid further research in the musicology area.