Avaliação de performance de algoritmos de reconhecimento de acordes musicais.

Abstract

In the musical context, chord progressions are capable of capturing several high-level characteristics of a song, such as possible associated feelings, genre, music tonality, rhythm, etc. However, identifying chord progressions is a process that requires the work of specialists, and it is a process that is not feasible to be done manually in large databases. Thus, the need arose to design classifiers that could extract this information from a sound signal. The objective of this work is to evaluate the performance of algorithms aimed at this purpose that are publicly available, verifying which learning strategies used in the algorithms are most promising, as well as identifying the biggest gaps in these models, in order to guide future efforts in the creation of new classifiers and training datasets.