Automatic Music Genre Classification - Supervised Learning Approach
Tulisalmi-Eskola, Johanna (2022)
2022
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202201031034
https://urn.fi/URN:NBN:fi:amk-202201031034
Tiivistelmä
Modern digital music libraries are huge. Searching and retrieving requested piece of music is challenging. Genre is used for identifying music. A computer utilising Machine Learning (ML) model can automatically classify a piece of music into corresponding music genre such as classical, jazz, or pop.
In this study a selection of conventional machine learning music genre classifier models were created and ranked by accuracy. In addition, a study on manual music genre classification was conducted. Performance of created models were compared against the state-of-art model and manual classification. Music content-based audio features were used as input data to models. This contrasted with models that use album, artist, social media, or symbolic representation of music as input.
Results of the study show that the performance accuracy of these created models varied from 64.03 % to 80.00 %. Support Vector Machine classifier being the most accurate and slightly less accurate compared to accuracy of MIREX state-of-art model (83.55 %) and manual classification accuracy (90.00 %) obtained from manual classification task included in this study.
In this study a selection of conventional machine learning music genre classifier models were created and ranked by accuracy. In addition, a study on manual music genre classification was conducted. Performance of created models were compared against the state-of-art model and manual classification. Music content-based audio features were used as input data to models. This contrasted with models that use album, artist, social media, or symbolic representation of music as input.
Results of the study show that the performance accuracy of these created models varied from 64.03 % to 80.00 %. Support Vector Machine classifier being the most accurate and slightly less accurate compared to accuracy of MIREX state-of-art model (83.55 %) and manual classification accuracy (90.00 %) obtained from manual classification task included in this study.