Resumen de Smart Song Equalization Based on the Classification of Musical Genres
Jesus Iriz, Miguel Ángel Patricio Guisado, José Manuel Molina López, Antonio Berlanga de Jesús
Music equalization is the process of trimming or raising specific frequencies (or a range of frequencies, called “frequency bands”) to increase the quality of a song. This paper presents an architecture that obtains a smart song equalization for a song based on a prediction of its musical genre. This prediction is based on a neural model designed in this work for the classification of musical genres in separated segments of a song. Each song needs its own equalizer setting that will not be the same throughout the song. For each segment, a distribution of probabilities of belonging to different genres is calculated, and then the song is equalized based on this distribution and different configuration options such as genres taken into account and using interpolation for the segments or not. These configurations are discussed in the paper. Both automatic music equalization and song genre classification are poorly researched matters, a gap we try to cover by investigating previous works in the area of study. Different proofs of concept are presented in the work to show the operation of the automatic equalizer. Adapting the solution for a smart equalizer can produce a system capable of automatically improving the music millions of people listen to daily in their mobile devices or applications such as Spotify and Youtube.
