Automatic classification of instruments from supervised methods of machine learning

Sorting instruments is not an easy task for humans or computers, especially when it comes to elements with the same acoustic properties, such as wind, percussion, or strings. Nevertheless, the use of audio descriptors and artificial intelligence techniques can make this duty more accessible. In this paper, three supervised methods, Naive Bayes, decision tree and Support Vector Classifier (SVC) are used to categorize acoustic guitar and bass sounds in a database, using as a parameter the information extracted from audio descriptors. The research resulted in a performance comparison of these three algorithms, considering their hit rates and processing time when classifying samples in different parts of the dataset. After all, some relevant considerations about the feasibility of automatically classifying instruments are presented.