Computational music: Analysis of music forms

With the development of computational science, many fields, including computational linguistics (sequence processing) and computational vision (image processing), have enabled various applications and automation with satisfactory results. However, the development of Computational Music Analysis (CMA) is still in its infancy. The main factor hindering the development of CMA is the complex form found in music pieces, which can be studied and analyzed in many different ways. Considering the advantages of Deep Learning (DL), this paper envisions a methodology for using DL to promote the development of Music Form Analysis (MFA). First, we review some common music forms and emphasize the significance and complexity of music forms. Next, we overview the CMA in two different processing ways, i.e., sequence-based processing and image-based processing. We then revisit the aims of CMA and propose the analysis principles that need to be satisfied for achieving the new aims during music analysis, including MFA. Subsequently, we use the fugue form as an example to verify the feasibility and potential of our envisioned methodology. The results validate the potential of using DL to obtain better MFA results. Finally, the problems and challenges of applying DL in MFA are identified and concluded into two categories, namely, the music and the non-music category, for future studies.