Query by Humming for Song Identification Using Voice Isolation

Alfaro-Paredes E.; Alfaro-Carrasco L.; Ugarte W. · Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (2021) · DOI: 10.1007/978-3-030-79463-7_27

Resumen

There are some methods for searching in large music databases, like searching by song name or artist name. However, in some cases these methods are not enough. For instance, a person might not remember the name of a song, but might remember its melody. In Music Information Retrieval, there is a task called Query-By-Humming, which allows retrieving a rank of songs that are similar to an audio humming. In this research, we propose the use of vocal isolation methods to improve query-by-humming systems. To achieve this, different configurations of Query-by-Humming systems were tested to analyze the results and determine in which cases our proposal works better. The results showed that vocal isolation improves the performance of Query-by-Humming systems when the music collection consists of modern songs.

Instituciones: Universidad Peruana de Ciencias Aplicadas