Klasifikasi Genre Dan Subgenre Musik Berbasis Fitur Mel-Frequency Cepstral Coefficients Menggunakan Convolutional Neural Network

Fahreza, Dimas Iqbal (2022) Klasifikasi Genre Dan Subgenre Musik Berbasis Fitur Mel-Frequency Cepstral Coefficients Menggunakan Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07211840000045-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (3MB)

Abstract

Dewasa ini, dengan semakin banyaknya musik yang diunggah ke internet, serta perkembangan platform untuk mendengarkan musik secara daring, seperti Spotify, Bandcamp, Deezer, SoundCloud, dan lain sebagainya, pengguna dapat mendengarkan musik yang mereka sukai dimana saja dan kapan saja. Jumlah musik yang sangat banyak tersebut perlu untuk dikategorikan berdasarkan genre dan subgenrenya secara otomatis karena sangatlah tidak mungkin bagi manusia untuk melakukannya secara manual. Selain itu, kurangnya penelitian mengenai klasifikasi subgenre pada ranah Music Information Retrieval (MIR) menjadi salah satu motivasi. Penelitian ini menguji apabila algoritma Convolutional Neural Network (CNN) dapat digunakan untuk mengklasifikasikan subgenre musik dengan cara menampilkan tiga subgenre teratas dengan menggunakan Mel-Frequency Coefficient of Cepstrum (MFCC) dari musik yang didapat pada dataset Free Music Archive (FMA) sebagai fitur utamanya. Pengujian yang dilakukan adalah menggunakan parameter Batch Size dan Epoch yang berbeda-beda, serta melakukan simulasi untuk mengetahui pengaruh durasi terhadap akurasi. Akurasi tertinggi menggunakan testing data didapat pada Batch Size = 32, dan Epoch = 100 dengan hasil 46,39%. Durasi telah terbukti memiliki peran besar dalam klasifikasi. Dengan menggunakan dataset simulasi, telah dilakukan simulasi untuk menampilkan tiga subgenre dengan probabilitas tertinggi, dan didapatkan akurasi tertinggi pada masukan musik dengan durasi 30 detik, yaitu dengan range akurasi antara terendah dan tertingginya sebesar 70,33% sampai dengan 73,22%. Selain itu, didapat bahwa akurasi total seiring bertambahnya durasi tidak pernah mengalami penurunan.
==============================================================================================================================
Within the increase of music uploaded to the internet, also the development of online music platforms such as Spotify, Bandcamp, Deezer, SoundCloud, etc. as of lately, users can enjoy their favorite music and artists anytime anywhere. Such amount of music have to be classified to their own genres and subgenres automatically because it’s impossible for humans to do such thing manually. In addition, the amount of research on music subgenre classification in the field of Music Information Retrieval (MIR) is still lacking. This research examines the efficiency of Convolutional Neural Network (CNN) architecture at classifying music subgenre by measuring the system’s performance at showing top three subgenres with the highest probabilities using Mel-Frequency Coefficient of Cepstrum (MFCC) extracted from track input provided by Free Music Archive (FMA) as its feature. Various amount of Batch Size and Epoch were used for testing purposes. Other than that, simulation to see how duration would affect the performance was also done. The highest accuracy using testing data was achieved using Batch Size = 32, and Epoch = 100 with the result of 46.39%. Based on the simulation, duration has been proven to have impact on the system’s performance. The simulation was done using simulation dataset by showing top three subgenres with the highest probability and the highest accuracy was achieved at track inputs with 30s duration, with the result of a range between its lowest and highest which is 70.33% up to 73.22%. Lastly, it was found that as the duration increases, the total accuracy never dropped.

Item Type:	Thesis (Other)
Additional Information:	RSKom 005.74 Fah k-1 2022
Uncontrolled Keywords:	Klasifikasi, Subgenre Musik, Mel-Frequency Coefficient of Cepstrum, Convolutional Neural Network. Classification, Music Subgenre, Mel-Frequency Coefficient of Cepstrum, Convolutional Neural Network.
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science. EDP
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	17 Jun 2026 03:07
Last Modified:	17 Jun 2026 03:07
URI:	http://repository.its.ac.id/id/eprint/133843

Actions (login required)

View Item