Orin: The Nigerian Music Dataset

Music Information Retrieval (MIR) is the task of extracting higher-level information such as genre, artist or instrumentation from music [1]. Music genre classification is an important area of MIR and a rapidly evolving research area. Presently, slight research work has been done on automatic music genre classification of Nigerian songs. Hence, this study presents a new music dataset named ORIN which mainly cores traditional Nigerian songs of four genres (Fuji, Juju, Highlife and Apala). The ORIN dataset consists of 208 Nigerian songs downloaded from the internet. Timbral Texture Features were then mined from one or two 30 second segments from each song using the Librosa [2] python library. Songs features were mined straight from the digital song files. Each piece of song was sampled at 22.5Khz 16-bit mono audio files. The song signal was then shared into frames of 1024 samples with 50% overlay between successive frames. A Hamming window is applied without pre-emphasis for each frame. Then, 29 averaged Spectral energies are obtained from a bank of 29 Mel triangular filters followed by a DCT, yielding 20 Mel frequency Cepstrum Coefficients (MFCC). The mean and standard deviation of the values taken across frames is considered as the representative final feature that is fed to the model for each of the spectral features. These features consist of the time (FFT) and frequency (MFCC) feature sets of the dataset domains.