Soundscape Emotion Recognition Models

Neural Networks for predicting Russell's valence and arousal circumplex model of affect for soundscapes.

---

Feature Set Dimensions

• Database feature set dimension: 1213 x 122
• Reduced feature set dimension: 1213 x 74

---

Trained Neural Network

	Arousal	Valence
R Squared	0.8684	0.7281
Mean Squared Error	0.0438	0.0906

---

SVR baseline model (122 x 1)

Arousal	Mean	Std
R Squared	0.793712	0.031416
Mean Squared Error	0.067285	0.009299

Valence	Mean	Std
R Squared	0.562547	0.048298
Mean Squared Error	0.143315	0.014217

SVR reduced feature model (74 x 1)

Arousal	Mean	Std
R Squared	0.785556	0.026073
Mean Squared Error	0.071068	0.009651

Valence	Mean	Std
R Squared	0.547271	0.051466
Mean Squared Error	0.148914	0.011666

---

Neural Network (122 x 1)

Arousal	Mean	Std
R Squared	0.722813	0.235087
Mean Squared Error	0.068943	0.033179

Valence	Mean	Std
R Squared	0.494576	0.173223
Mean Squared Error	0.137190	0.030402

Neural Network reduced feature (74 x 1)

Arousal	Mean	Std
R Squared	0.730940	0.161783
Mean Squared Error	0.070679	0.028536

Valence	Mean	Std
R Squared	0.504163	0.121443
Mean Squared Error	0.138414	0.038270

---

Reference

• Database from Emo-Soundscapes: https://metacreation.net/emo-soundscapes/