[Paper Review] A Categorical Approach for Recognizing Emotional Effects of Music

Mohsen Sahraei Ardakani, Amir Arbabi|arXiv (Cornell University)|Sep 17, 2017

Music and Audio Processing20 references3 citations

TL;DR

This paper proposes a categorical approach to recognizing emotional effects in music using a combination of audio features and machine learning. It applies Fisher's linear discriminant analysis to classify six emotional labels—happy, sad, relaxing, exciting, epic, and thriller—extracting temporal, spectral, and energy features, with Support Vector Machines achieving 77.4% accuracy in identifying 'epic' music parts as the highest-performing category.

ABSTRACT

Recently, digital music libraries have been developed and can be plainly accessed. Latest research showed that current organization and retrieval of music tracks based on album information are inefficient. Moreover, they demonstrated that people use emotion tags for music tracks in order to search and retrieve them. In this paper, we discuss separability of a set of emotional labels, proposed in the categorical emotion expression, using Fisher's separation theorem. We determine a set of adjectives to tag music parts: happy, sad, relaxing, exciting, epic and thriller. Temporal, frequency and energy features have been extracted from the music parts. It could be seen that the maximum separability within the extracted features occurs between relaxing and epic music parts. Finally, we have trained a classifier using Support Vector Machines to automatically recognize and generate emotional labels for a music part. Accuracy for recognizing each label has been calculated; where the results show that epic music can be recognized more accurately (77.4%), comparing to the other types of music.

Motivation & Objective

• To address the inefficiency of current music retrieval systems based on album metadata.
• To explore the feasibility of automatically tagging music with emotional labels based on audio content.
• To evaluate the separability of emotional categories using statistical feature analysis.
• To develop a machine learning model capable of recognizing emotional effects in music segments.
• To identify which emotional categories are most distinguishable using audio features.

Proposed method

• Extracted temporal, frequency, and energy-based audio features from music segments.
• Applied Fisher's linear discriminant analysis to assess separability between emotional labels.
• Selected six emotional adjectives—happy, sad, relaxing, exciting, epic, thriller—as target labels.
• Trained a Support Vector Machine (SVM) classifier on the feature set to predict emotional categories.
• Used the Fisher separation theorem to evaluate the discriminative power of feature sets across emotional classes.
• Optimized classification performance by analyzing feature distributions across emotional categories.

Experimental results

Research questions

• RQ1 Can emotional labels in music be effectively separated using audio features?
• RQ2 Which emotional categories exhibit the highest separability in the feature space?
• RQ3 How accurately can a machine learning model classify music into predefined emotional categories?
• RQ4 Which emotional label is most reliably recognized by the proposed system?
• RQ5 What is the contribution of different audio features (temporal, spectral, energy) to emotional classification?

Key findings

• The maximum separability between emotional categories was observed between 'relaxing' and 'epic' music parts.
• The Support Vector Machine classifier achieved the highest accuracy of 77.4% in recognizing 'epic' music segments.
• The 'epic' category was the most accurately classified emotional label among the six.
• Temporal, frequency, and energy features collectively contributed to effective emotional classification.
• The results support the feasibility of using audio features and statistical learning for automatic emotional tagging of music.
• The study demonstrates that emotional categorization of music can be systematically approached using Fisher’s discriminant analysis.