Noise pollution is a critical aspect of environmental quality that significantly impacts human well-being. This study explored the EmoSynth dataset, comprising 144 isolated synthetic periodic waveform audio files labeled by multiple individuals for perceived emotions based on Valence and Arousal. The defined approach involved reorganizing the Valence and Arousal values into three balanced classes and extracting the Gammatone Cepstral Coefficients (GTCCs) from each recording. These coefficients aim to mimic human perception of sounds, making them suitable for describing how sound might affect individuals. Using these features, a two-phase classification task was established to analyze the emotional content of the audio files. In the initial phase, employing a Bagged Trees classifier, the three levels of Arousal and Valence were the output targets. The model achieved validation accuracies of 74.5% for Arousal and 77.8% for Valence, and maintained high accuracies in the testing phase, reaching values of 73.3% and 83.3% respectively. Subsequently, in the second phase, a multi-output classification using a Random Forest model was conducted, successfully identifying both Arousal and Valence labels for 88.5% of the test instances. These findings contribute to understanding how soundscapes influence emotional experiences, productivity, and well-being. Furthermore, these results underscore the importance of considering environmental noise in assessments of overall quality of life and psychological health.
Exploring The Impact of Soundscapes on Emotional States: A Data Driven Approach on Environmental Noise Recording
Di Cesare M. G.Primo
;Perpetuini D.;Cardone D.;Merla A.Ultimo
2024-01-01
Abstract
Noise pollution is a critical aspect of environmental quality that significantly impacts human well-being. This study explored the EmoSynth dataset, comprising 144 isolated synthetic periodic waveform audio files labeled by multiple individuals for perceived emotions based on Valence and Arousal. The defined approach involved reorganizing the Valence and Arousal values into three balanced classes and extracting the Gammatone Cepstral Coefficients (GTCCs) from each recording. These coefficients aim to mimic human perception of sounds, making them suitable for describing how sound might affect individuals. Using these features, a two-phase classification task was established to analyze the emotional content of the audio files. In the initial phase, employing a Bagged Trees classifier, the three levels of Arousal and Valence were the output targets. The model achieved validation accuracies of 74.5% for Arousal and 77.8% for Valence, and maintained high accuracies in the testing phase, reaching values of 73.3% and 83.3% respectively. Subsequently, in the second phase, a multi-output classification using a Random Forest model was conducted, successfully identifying both Arousal and Valence labels for 88.5% of the test instances. These findings contribute to understanding how soundscapes influence emotional experiences, productivity, and well-being. Furthermore, these results underscore the importance of considering environmental noise in assessments of overall quality of life and psychological health.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
