Melodic Pattern Recognition and Similarity Modelling: A Systematic Survey in Music Computing

Musical pattern identification is crucial for various classification and retrieval applications in computational musicology. Feature learning is the basic task, and features act as a basis for the Pattern Recognition (PR). Selecting an appropriate approach is vital to the accuracy of the retrieval algorithms. This research gives a comprehensive review of approaches used for PR and similarity modelling. It systematically analyses various approaches for melodic feature identification and comparatively evaluates the work done in the literature in terms of software tools used, melodic pattern representations, and matching. The study discusses the benefits and limitations of various approaches along with the challenges to be addressed for melodic PR. Results show a wide variety of approaches for different music genres and applications. Further, analysis shows that statistical and symbolic approaches were being used predominantly, and deep learning approaches are gaining popularity in recent times.