The Perception of Emotion in the Singing Voice

Transcription

1 The Perception of Emotion in the Singing Voice Emilia Parada-Cabaleiro 1,2, Alice Baird 1,2, Anton Batliner 1,2, Nicholas Cummins 1,2, Simone Hantke 1,2,3, Björn Schuller 1,2,4 1 Chair of Embedded Intelligence for Health Care and Wellbeing, Augsburg University, Germany 2 Chair of Complex and Intelligent Systems, University of Passau, Germany 3 MISP Group, MKK, Technische Universität München, Germany 4 GLAM - Group on Language, Audio & Music, Imperial College London, UK This work was supported by the European Union's Seventh Framework and Horizon 2020 Programmes under grant agreements No (ERC StG ihearu) and No (RIA DE-ENIGMA).

2 Music can induce emotions In a digital world emotional content is an ever increasing feature music retrieval Web emotional playlist as Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 2

3 Which kind of music evokes emotions? ORCHESTRATION? trumpet, xylophone ARTICULATI ON? DYNAMICS? >, stacc, sfz mf, ppp, ff Rhythm? binary, syncopated HARMONY? tonal, modal, jazz Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 3

4 Voice as communication channel Speech Singing The capacity to express emotion Rethoric of ancient Greece and Rome: Cicero, Aristotle Work songs Religious chant Lullaby Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 4

5 The study of emotions in singing SYSTEM S FOR AUTOMATIC SPEECH EMOTION RECOGNITION LISTENING RETRIEVING ASSISTANCE DIGITAL ORGANIZATION LIBRARIES ONLY ACOUSTIC FEATURES MUSIC AL SINGING VOICE Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 5

6 Considering that The singing voice can naturally express emotions Music retrieval by emotional content is ever increasing The Musical features that relate to emotion are still unknown for automatic recognition of emotions in singing voice Research goal To evaluate which musical features of the singing voice relate to listeners perception of emotions Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 6

7 Methodology: Listening test Theory of emotions Categorical Dimensional Models for MIR (e. g., MIREX 5-cluster model) Perception test Bi-dimensional (more suitable for music evaluation) 7 level rating scale ihearu-play crowdsourcing platform Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 7

8 Methodology: Dataset Electronic manipulation in speech Random Splicing, Band Pass filtering Italian Folk Dataset 6 Italian folk songs (Canzone Romana in Roman dialect) Random splicing 0.5sec per segment (to disrupt Rhythmic-melodic Contour) Reversing (to disrupt Musical Syntax) Global tempo manipulation 25% slower, 50% faster (to disrupt Tempo) Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 8

9 Overview of the study 104 sung chunks (52 for each gender) Lasting 11.8 sec average (sd of 2.9 sec) Sung by 6 singers (3 for each gender) 26 not manipulated, 26 random-spliced, 26 reversed, 26 global tempo manipulated 24 listeners (18-30 years, sd of 3.5 years) 13 German, 11 non German Bi-Dimensional test (arousal/valence, 7 level) None of them Italian speaker. Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 9

10 Results p (sig) Clear / random splicing Clear / reversing Clear / tempo manipulation dimension Mean differences between clean and manipulated signals (ANOVA ) Starred results indicate p <.05 in Tukey s post hoc test Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 10

11 Results Clean/random splicing valence (f/m) Clean/reversing Clean/tempo manipulation Mean differences between clean and manipulated signals for each singer. ID1, ID2, ID3 (female); ID4, ID5, ID6 (male); ID1 and ID5 (Allegro). Starred results indicate p <.05 in Tukey s post hoc test Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 11

12 Results IN Indian TI Tunesian + Iranian EU British + Spanish Clean/random splicing Clean/reversing valence (f/m) Clean/tempo manipulation valence (f/m) Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 12

13 Conclusions The perception of emotions in Italian folk seems to be less influenced by manipulation techniques in German listeners. Confirming previous studies, culture may influence the perception of emotion in music. Rhythmic-melodic contour is linked to both dimensions. Musical syntax and allegro Tempo (fast speed) has shown to be specially related to Valence. Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 13

14 Future Work Further perception tests - larger listener groups from a variety of musical traditions. Emotional models for music connect our work with MIR and MDL studies. Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 14

15 This work was supported by the European Union s Seventh Framework and Horizon 2020 Programmes under grant agreement No (ERC StG ihearu) and No (RIA DE-ENIGMA) ihearu.eu Questions? de-enigma.eu For further information: emilia.parada-cabaleiro@informatik.uni-augsburg.de Parada-Cabaleiro et al. The Perception of Emotion in the Singing Voice 15