EE 473
DIGITAL SIGNAL PROCESSING
Term Project
CONTENT-BASED CLASSIFICATION
OF MUSICAL INSTRUMENT SOUNDS USING GAUSSIAN
METHODS
Project Members
Murat Aksoy
Hasan Ayaz
Ender Konukoglu
Problem Statement
The objective of this project is to classify musical instrument sounds using their cepstral coefficients. The selection is made using the pre-obtained sample set in a database consisting of several notes from various musical instruments.
Introduction
Classification of audio signals according to their content has been a major concern in recent years. There have been many studies on audio content analysis, using different features and different methods. It is a well known fact that audio signals are baseband, one-dimensional signals. General audio consists of a wide range of sound phenomena such as music, sound effects, environmental sounds, speech and nonspeech signals.
In this project we are mainly concerned with the classification of
audio signals that are sampled and recorded from different kinds of musical
instruments. The classification of musical instrument sounds, at first step,
requires the extraction of certain features related to the input sound sample,
which may include root-mean-square amplitude envelope, constant Q transform
frequency spectrum, Multidimensional Analysis Scaling trajectories, cepstral coefficients, spectral centroid and presence of
vibrato[1].
There are two main approaches to this problem of content based
classification based on previous extracted features :
The first which uses deterministic
methods and the one that utilizes probabilistic techniques.
There are many research efforts, high accuracy audio classification
is only achieved for the simple cases such as speech/music discrimination. Pfeiffer et al ,
presented a theoretic framework and application of automatic audio content
analysis using some perceptual features.
On the other hand, Saunders presented a speech/music classifier based on
simple features such as zero crossing rate and short time energy for radio
broadcast.
Scheirer et al
conducted many experiments with different classification models including GMM
(Gaussian Mixture Model), BP-ANN (Back Propagation Artificial Neural Network)
and KNN (K-Nearest Neighbour). Many
other works have been done to enhance audio classification algorithms such as
pre-classification of audio recordings into speech, silence, laughter and
non-speech sounds, in order to segment discussion recordings in meetings. The
usage of taxonomic structures also helps to enhance classification performance. In the work by Zhang and Kuo,
pitch tracking methods are introduced to discriminate audio recordings into
more classes, such as songs, speeches over music, with a heuristic-based model.
Accuracy of above 90% is reported. Srinivasan et al, try to detect and classify audio that
consists of mixed classes, such as combinations of speech and music together
with background sound. The accuracy of classification is over 80% [2]
Methodology
The project makes use of the probabilistic
methods and is planned to put forward algorithms to classify musical
instruments via their cepstral characteristics. As
the first step, the cepstral coefficients from the
pre-obtained database are to be obtained by first taking FFT of the signal,
then taking the logarithm, and re-taking the inverse FFT. A mean and covariance
matrix will then be extracted from this information and fit into a Gaussian
function, which will be the main tool to be used in the identification of the
input signal.
In order to compare the cepstral
characteristics, the cepstral coefficients from the
input sample are also extracted and this value is put into the Gaussian
distribution stated above, and decision-making is accomplished using Bayesian
analysis. The Bayesian decision is made by determining the class whose Gaussian
distribution results in the highest probability by this input pattern vector.
Figure 1. Flow Chart of the classification algorithm
Time Schedule
|
Database searching and determination of the instruments to be worked
on |
7 days |
|
Creating the cepstral coefficients matrix for each sample in the
database and fitting into appropriate Gaussian functions |
7 days |
|
Implementation
of Bayesian Decision algorithm |
15 days |
|
Testing the
performance of the algorithm by using pre-recorded database samples |
10 days |
|
Enhancements on
the overall algorithm:
(pitch, brightness, etc.)
|
10 days
(optional) |
References
[1] Multi-feature
Musical Instrument Sound Classifier w/user determined generalisation performance,Ian Kaminskyj,Electrical & Computer Systems
Engineering,
[2] A Robust Audio Classification and Segmentation Method
Lie Lu, Hao Jiang and Hong-Jiang Zhang
Microsoft research,
[3] Indexing
Audio Databases with Musical Information; Alicja A. Wieczorkowska, Polish-Japanese Institute of Information
Technology; Zbigniew W. Ras,
[4] Sound
Identification and Direction Detection in MATLAB for Surveillance Applications,
M. Cowling, R. Sitte, Griffith University Faculty of Engineering and
Information Technology, Queensland,
Australia
[5] Automatic Sound Classification Inspired by
Auditory Scene Analysis
Silvia
Allegro, Michael Büchler, Stefan Launer
Signal Processing Department, Phonak AG, Switzerland + Department of Otorhinolaryngology,
University Hospital Zurich, Switzerland
[6] Julus T. Tou, Rafael C. Gonzalez,
Pattern Recognition Principles, Addison-Wesley Publishing Company
, 1974
[7]
Richard O. Duda, Peter E. Hart, Pattern
Classification and Scene Analysis, John Wiley & Sons Inc, 1973