MUSIC EMOTION DETECTION USING MULTI-LEVEL CLASSIFICATION

CHAPTER ONE

1.0 Introduction

This chapter introduces the proposal of the study. Sections 1.1 and 1.2 presents the background to

the study and statement of the problem, Sections 1.3 and 1.4 presents the motivation behind the

study and specific objectives of the study, while Sections 1.5, 1.6 and 1.7 presents the research

methodology, expected contribution and scope of the study respectively. Lastly, section 1.8

presents the definition of key terms.

1.1 Background to the Study

The concept music has been seen or defined by different scholars, but what impresses most, is that

music is seen generally as an organized sound that is pleasant to the ear. This simply means that

any sound that is arranged in a haphazard and incoherent way or manner is regarded as a bundle

or noise or meaningless (Chauhan and Desai, 2018). Music Emotion Recognition (MER) refers to

the task of finding a relationship between music and human emotions. Nowadays, this type of

analysis is becoming more and more popular, music streaming providers are finding very helpful

to present users with musical collections organized according to their feelings.

According to Chien et al. (2020), Music Emotion Recognition was proposed for the first time in

the Music Information Retrieval (MIR) community in 2007, during the annual Music Information

Research Evaluation exchange. Audio and lyrics represent the two main sources from which it is

possible to obtain low and high-level features that can accurately describe human moods and

emotions perceived while listening to music. An equivalent task can be performed in the area of

Natural Language Processing (NLP) analyzing the text information of a song by labeling a

sentence, in a case one or more lyrics lines, with the emotion associated to what it expresses. A

typical music emotion detection approach consists in training using the lyrics.
In Cristianini (2018), Music is not only for entertainment or pleasure, social and psychological

effects of music have been studied extensively for decades. For centuries now, the world is

witnessing ever-increasing growth of the on-line music. Efficient and accurate automatic music

information processing (in particular, accessing and retrieval) will be extremely important research

issues of this century. Traditionally musical information has been retrieved and/or classified based

on standard reference information, such as the name of the composer, the title of the work, the

album title, the style of music, and so on. These basic pieces of information will remain essential,

but information retrieval based purely on these is far from satisfactory. Music is a branch of art in

which emotions are expressed with sounds (Davies and Pumbley, 2017). Due to the variety and

richness of music content, it has been the subject of research in many studies in different fields.

Music emotion classification systems may be utilized actively for a variety of reasons, including

organizing personal music collections and constructing repertoire recommendation systems, music

therapy, and emotional illness treatment (Delbouys et al. 2018). Recent advances in the study of

the brain have enables enhancement in the understanding of the way that active engagement with

music influences other development. Although the knowledge of the way the brain works is still

in its infancy some of the fundamental processes involved in learning have been established.

1.2 Statement of Problem

The significance of music on an individual's emotions has been generally acknowledged. After the

day’s toils and hard work, both the primitive and modern man are able to relax and ease in the

melody of the music. Measuring and assigning emotions to music is not a straightforward task:

people face the difficulty of song selection, especially songs that match individuals’ current

emotions. Looking at the long lists of unsorted music, individuals will feel more demotivated to
look for the songs they want to listen to, the sentiment/mood associated with a song can be derived

by a combination of many features, moreover, emotions expressed by a musical excerpt and by its

corresponding lyrics do not always match, also, the annotation of mood tags to a song turns out to

be highly changing over the song duration and therefore one song can be associated with more

than one emotion. There happens to be no unified emotion representation in the field of MER,

there is no consensus upon which and how many labels are used and if emotions should be

considered as categorical or continuous.

1.3 Research Motivation

Most of the work done in the field of music signal processing, it covers applications such as

classifying musical genre, determining the musical instruments used, and revealing musical

similarities such as notes and chords. Recently, there has also been increased interest in studies

aimed at automatically analyzing and recognizing the emotional content of a piece of music

however, the efficiency and performance is still at the low side.

1.4 Aim and Specific Objectives of the Study

This research work aims to classify the emotional content of music using a classification model

The specific objectives are to:

a. Data acquisition (secondary data)

b. Use a classification model for music emotion using supervised machine learning technique.

c. Validate and evaluate the model (b)

1.5 Methodology
This study adopts a multilevel machine learning approach, using random forest and support vector

machine. For the aim of this research work to be achieved, the following major procedures and

steps shall be taken.

i. Data collection: this implies the gathering of vital samples of data relevant to the subject

domain of the research work. note that: data can be primary or secondary i.e collected using

questionnaires, online web application information gathering platforms and so many more,

or downloading ready-made / raw data from an online server.

ii. Data preprocessing: After all required data has been collected and stored, it must be

preprocessed. The preprocessing may refer to the normalization and/or discretization of

some parameters in a given range; or generating new parameters that did not exist in the

original database.

iii. Data analysis: Data collected must be inspected, cleaned, transformed, and modeled to

discover useful information, informing conclusions, and support decision-making.

iv. Model Building: Machine learning provides numerous algorithms to be applied to data to

perform prediction task. Each algorithm works effectively depending on the data collected.

In this stage, random forest is the models to be applied.

v. Model Training: In this stage, the random forest and support vector model would be

trained with a certain percentage of the dataset specified for training.

vi. Model Evaluation: After the model had been trained, the model would be evaluated and

the precision, recall, and accuracy would specify the performance of the model.

vii. Prediction: The output of the tested data would be generated into a file.

1.5.1 Conceptual Model

 Data
Normalizatio
n Classification
Algorithm
Music data Train Model and
Data Preparation
Test Model

Data
Cleaning
Cleaningsd

Evaluate Model Prediction

Output

FIGURE 1.1 System architecture for proposed model

1.6 Significance / Expected contribution of the Study

This research work is expected to establish a solid foundation for music emotion detection by

exploring a multilevel classification technique. It is also expected that, thanks to the dataset and
methods to be presented in this study, it will contribute to the creation of more collaborative work

in the field of computing and music with datasets consisting various samples

1.7 Scope of the Study

Music emotion are a great indicator of the state of a music for a person. Indeed, the most natural

way to express emotions is through facial expressions, however, people express their emotions

through music and song snippets. Humans tend to link the music they listen to, to the emotion they

are feeling. This works covers the area of detecting the emotion behind a music using a multi-level

classification approach

1.8 Definition of Terms

Machine learning: an application of artificial intelligence which gives a machine the ability to

learn over time and improve based on past experiences without being explicitly programmed

Supervised machine learning: a branch of machine learning which makes use of labeled data to

predict the result of the next input

Classification: this is a technique in supervised learning and data mining which are capable of

processing a large amount of data.

Support vector machine: this is a type of classification model with an associated learning

algorithm that analyze data for classification and regression analysis.

Random Forest algorithm: this is a powerful ensemble learning technique for classification that

operates by constructing a multitude of decision trees at training time

Algorithm: a procedure or formula for solving a problem, based on conducting a sequence of

specified actions

Model: a simplified representation used to explain the working of a real-world system or event

Prediction: a forecast statement of what will happen in the future