©2017, IJCERT All Rights Reserved Page |43
Volume 4, Issue 2, February-2017, pp. 43-47 ISSN (O): 2349-7084
International Journal of Computer Engineering In Research Trends
Music Genre Classification Using MFCC, K-NN
and SVM Classifier
Nilesh M. Patil
1
, Dr. Milind U. Nemade
2
,
Ph.D Research Scholar
1
,
Pacific Academy of Higher Education and Research University, Udaipur, India.
Email ID: nileshdeep@gmail.com.
Professor & Head
2
,
Department of Electronics Engineering, K J Somaiya Institute of Engineering & Information Technology, Mumbai.
Email ID: munemade@gmail.com
--------------------------------------------------------------------------------------------------------------------------------
Abstract:-The audio corpus available today on Internet and Digital Libraries is increasing rapidly in huge
volume. We need to properly index them if we want to have access to these audio data. The search engines
available in market also find it challenging to classify and retrieve the audio files relevant to the user’s interest.
In this paper, we describe an automated classification system model for music genres. We firstly found good
feature for each music genre. To obtain feature vectors for the classifiers from the GTZAN genre dataset,
features like MFCC vector, chroma frequencies, spectral roll-off, spectral centroid, zero-crossing rate were
used. Different classifiers were trained and used to classify, each yielding varying degrees of accuracy in
prediction.
Keywords: Music, MFCC, K-NN, SVM, GTZAN dataset.
-----------------------------------------------------------------------------------------------------------------------------------------------
1. Introduction:
Music classification is an interesting problem
with varying applications from Drinkify to Pandora. Music
classification is still considered to be one of the research
area due to the challenge in selection and extraction of
optimal audio features. Music genre classification has been
a challenging task in the field of Music Information
Retrieval (MIR). Music genres are inherently subjective
due to which they are hard to systematically and
consistently describe. Genre classification, till now, had
been done manually by concatenating it to metadata
repository of audio files. This paper however aims at
content-based classification, focusing on information
within the audio. We used traditional machine learning
approach for classification by finding suitable features of
audio signals, training classifier on feature data and make
predictions..
2. Related Works
Tzanetakis and Cook [1] pioneered the work on
music genre classification using machine learning
technique. They created the GTZAN dataset and is to date
considered as a standard for genre classification.
Changsheng Xu et al. [2] have shown how to use support
vector machines (SVM) for this task. Authors used
supervised learning approaches for music genre
classification. Scaringella et al. [3] gives a comprehensive
survey of both features and classification techniques used
in the music genre classification. Riedmiller [4] used
unsupervised learning creating a dictionary of features.
3. Description
An open source software framework called
MARSYAS (Music Analysis, Retrieval, and Synthesis for
Audio Signals) is available for audio processing with
specific emphasis on Music Information Retrieval
Applications [6]. MARSYAS website gives access to
GTZAN dataset which is a collection of 900 audio tracks
each 30 seconds long. There are 9 genres represented, each
containing 100 tracks. All the tracks are 22050Hz Mono
Available online at: www.ijcert.org