Music Emotion Recognition with Standard and Melodic Audio Features

Abstract

We propose a novel approach to music emotion recognition by combining standard and melodic features extracted directly from audio. To this end, a new audio dataset organized similarly to the one used in MIREX mood task comparison was created. From the data, 253 standard and 98 melodic features are extracted and used with several supervised learning techniques. Results show that, generally, melodic features perform better than standard audio. The best result, 64% f-measure, with only 11 features (9 melodic and 2 standard), was obtained with ReliefF feature selection and Support Vector Machines.

Notes

¹ http://www.allmusic.com/moods.

² http://www.music-ir.org/mirex/wiki/2013:Audio_Classification_%28Train/Test%29_Tasks#Audio_Mood_Classification.

³ http://www.ismir.net/.

⁴ http://www.music-ir.org/mirex/wiki/2010:MIREX2010_Results.

⁵ http://mir.dei.uc.pt/resources/MIREX-like_mood.zip.

⁶ http://psysound.wikidot.com/.

⁷ http://www.jyu.fi/hum/laitokset/musiikki/en/research/coe/materials/mirtoolbox.

⁸ http://marsyas.info/.

⁹ http://www.cs.waikato.ac.nz/ml/weka/.

Additional information

Funding

This work was supported by the MOODetector project (PTDC/EIA-EIA/102185/2008), financed by the Fundação para Ciência e a Tecnologia (FCT) and Programa Operacional Temático Factores de Competitividade (COMPETE) – Portugal, as well as the PhD Scholarship SFRH/BD/91523/2012, funded by the Fundação para Ciência e a Tecnologia (FCT), Programa Operacional Potencial Humano (POPH) and Fundo Social Europeu (FSE). This work was also supported by the RECARDI project (QREN 22997), funded by the Quadro de Referência Estratégica Nacional (QREN).