Rokkmuusika alastiilide klassifitseerimine tugivektormasinatega
Files
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Käesolev töö keskendub rokkmuusika alastiilide automaatsele klassifitseerimisele. Töö eesmärk on näha, kui edukalt on seda võimalik teha ning kas alastiilide klassifitseerimisel on tulevikuks potentsiaali. Ülesande lahendamiseks valiti tugivektormasinate meetod. Töös on antud ülevaade eraldatud tunnustest, kasutatud alastiili gruppidest ja tugivektormasinate tööpõhimõttest. Selle töö eesmärgil koostati muusikakorpus, mis koosnes viiest alastiilide grupist. Nendeks gruppideks olid: progressiivne rokk, punkrokk, metal-muusika, ekstreem-metal ja klassikaline rokkmuusika. Tööks kasutati 500 lugu, millest 400 olid kasutusel mudeli treenimiseks ja 100 testimiseks. Tunnuste eraldamiseks kasutati jAudio võimalusi ja klassifitseerimiseks kasutati Wekat. Suurimaks klassifitseerimise täpsuseks saavutati 71%. Kvartiilhaaret kasutades saavutati 74% täpsust.
This paper focuses on performing automatic genre classification using subgenres of rock music. The purpose of this paper is to see how well it can be done and whether subgenre classification has potential for the future. Suport vector machines were chosen for this task. Overviews of the extracted features, used genre groups, and the basic ideas behind support vector machines are presented. For the purpose of this work, a dataset of five different subgenre groups was constructed. The groups were as follows: progressive rock, punk rock, general metal, extreme metal, and general rock music. A total of 500 songs was used, of which 400 songs was used to train the model and 100 songs was use to test it. Features were extracted using jAudio and classification task was done with Weka. Highest result achieved was the classification acuracy of 71%. With the use of interquartile ranges the accuracy reached 74%.
This paper focuses on performing automatic genre classification using subgenres of rock music. The purpose of this paper is to see how well it can be done and whether subgenre classification has potential for the future. Suport vector machines were chosen for this task. Overviews of the extracted features, used genre groups, and the basic ideas behind support vector machines are presented. For the purpose of this work, a dataset of five different subgenre groups was constructed. The groups were as follows: progressive rock, punk rock, general metal, extreme metal, and general rock music. A total of 500 songs was used, of which 400 songs was used to train the model and 100 songs was use to test it. Features were extracted using jAudio and classification task was done with Weka. Highest result achieved was the classification acuracy of 71%. With the use of interquartile ranges the accuracy reached 74%.