Audio features dedicated to the detection and tracking of arousal and valence in musical compositions

Figures & data

Figure 1. A–V emotion plane – Russell's circumplex model (Russell, 1980).

Table 1. Amount of examples in quarters on the A–V emotion plane.

Quarter abbreviation	Arousal–Valence	Amount of examples
Q1	high–high	93
Q2	high–low	70
Q3	low–low	80
Q4	low–high	81

Table 2. R² and MAE obtained for SMOreg.

	Essentia				Marsyas
	Arousal		Valence		Arousal		Valence
		MAE		MAE		MAE		MAE
Before attribute selection	0.48	0.18	0.27	0.17	0.63	0.13	0.15	0.16
After attribute selection	0.79	0.09	0.58	0.10	0.73	0.11	0.25	0.14

Table 3. R² and MAE for arousal and valence obtained for combinations of feature sets.

	Arousal		Valence
Features set		MAE		MAE
L	0.74	0.10	0.49	0.12
R	0.68	0.11	0.15	0.15
T	0.53	0.14	0.48	0.12
L+R	0.79	0.09	0.40	0.12
L+T	0.74	0.10	0.56	0.10
R+T	0.74	0.11	0.52	0.11
All (L+R+T)	0.79	0.09	0.58	0.10

Table 4. Selected features used for building the arousal regressor.

Features from set L+R+T	Features from set L+R
Average Loudness (L)	Barkbands Kurtosis (L)
Barkbands Spread (L)	Dissonance (L)
Melbands Crest (L)	Erbbands Flatness (L)
Melbands Flatness (L)	Erbbands Skewness (L)
Melbands Kurtosis (L)	Melbands Crest (L)
Melbands Skewness (L)	Melbands Kurtosis (L)
Melbands Spread (L)	Silence Rate (L)
Spectral Energy (L)	Spectral Energy (L)
Spectral Entropy (L)	Spectral Entropy (L)
Spectral Flux (L)	Spectral Flux (L)
Spectral Kurtosis (L)	Spectral Rolloff (L)
Spectral Rolloff (L)	Spectral Skewness (L)
Spectral Skewness (L)	Beats Count (R)
BPM Histogram (R)	Beats Loudness (R)
Danceability (R)	Danceability (R)
Onset Rate (R)	Onset Rate (R)
Beats Loudness Band Ratio (R)	Beats Loudness Band Ratio (R)
Chords Strength (T)
Beats Loudness Band Ratio (R)
HPCP Entropy (T)
Key Strength (T)
Chords Histogram (T)

Table 5. Selected features used for building the valence regressor.

Features from set L+R+T	Features from set L+T
High Frequency Content (L)	Melbands Crest (L)
Melbands Kurtosis (L)	Melbands Spread (L)
Melbands Skewness (L)	Pitch Salience (L)
Spectral Energy (L)	Silence Rate (L)
Zero Crossing Rate (L)	Spectral Centroid (L)
GFCC (L)	Spectral Energy (L)
MFCC (L)	Spectral Spread (L)
Beats Loudness (R)	Zero Crossing Rate (L)
Onset Rate (R)	GFCC (L)
Beats Loudness Band Ratio (R)	MFCC (L)
Chords Strength (T)	Chords Strength (T)
HPCP Entropy (T)	HPCP Entropy (T)
Key Strength (T)	Key Strength (T)
Chords Histogram (T)	Key Scale (T)

Figure 2. A–V maps for the song Let It Be by Paul McCartney (The Beatles).

Figure 3. A–V maps for Piano Sonata No. 8 in C minor, Op. 13 (Pathetique), 2nd movement, by Ludwig van Beethoven.

Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39(6), 1161–1178. doi: 10.1037/h0077714

 Web of Science ®Google Scholar