Classification is among the most important topics in machine learning. There are abundant applications of classification problems in real life that human beings can solve with the help of Machine learning algorithms. An interesting application for classification problems is to classify a given audio file and identify the predominant musical instrument. Music data, compared to other types of data, is more complex since it often involves multi- dimensional feature extraction. An audio sound can be represented in the form of a wave, which then can be described in terms of different characteristics: amplitude, wave length, frequency, velocity, and time-period. Differ- ent combinations of characteristics can result in different features that uniquely belong to an audio file.
In this project, we aim to explore and apply various machine learning methods for classification of audio data in music. Our main goals were to explore different features that can distinguish the instruments, learn how to extract these features in the most meaningful way, and finally learn to recognize and classify the predominant instrument for a given music audio file using suitable Machine learning methods.
The results demonstrated that recognizing the predominant instrument of a song is feasible by taking machine learning approaches. While numerous audio features can be extracted from a music piece, only a few of them largely contribute to solving our problem and demand dedicate processing to maximize their impact in providing a satisfying solution.