Feature Extraction for Traditional Malay Musical Instruments Classification System

Abstract

Automatic musical instrument classification system deals with a large number of sound database and various types of features schemes. With the lack of data preprocessing, it might become invaluable asset that can impact the whole classification tasks. In handling an effective classification system, finding the best data sets with the best features schemes often a vital step in the data representation and feature extraction process. Thus, this study is conducted in order to investigate the impact of several factors that might affecting the classification accuracy such as audio length, segmented frame size and data sets size (for training and testing) towards Traditional Malay musical instruments sounds classification system. The perception-based and MFCC features schemes with total of 37 features was utilized in this study. Meanwhile, Multi-Layered Perceptrons classifier is employed to evaluate the modified data sets and extracted features schemes in terms of their classification performance. Results show that the highest accuracy of 99.57% was obtained from the best data sets with the combination of full features. It is also revealed that the identified factors had a significant role to the performance of classification accuracy. Hence, this study suggest that further feature analysis research is necessary for better solution in Traditional Malay musical instruments sounds classification system problem.

Extraction of feature vectors for analysis of musical instruments

“…Zero Crossing Rate (ZCR) & Root Mean Square (RMS) are representations of the signal in time domain [3], [6].…”

“…1) Spectral Flux: Spectral flux is the distance between the spectra of each successive frame and is related to the variation of spectrum over time [6], [7]. Spectral flux is calculated using (3).…”

“…Spectral bandwidth: This feature indicates the frequency range of a signal weighted by its spectrum, calculated using(6).…”

An Optimized Back Propagation Learning Algorithm with Adaptive Learning Rate

“…Despite the popularity and provide successful solutions, BP also known for some drawbacks. The main drawback of BP occur because it uses gradient descent (GD) learning which requires careful selection of parameters such as network topology, initial weights and biases, learning rate, activation function and value for the gain in the activation function [7]. Despite of all those drawbacks, the popularity and the ability of back propagating learning is still increasing.…”

Correlation and clusterisation of traditional Malay musical instrument sound using the I-KAZTM statistical signal analysis

“…It is a systematic approach that identifies and categorises the musical sound signals. Generally, the sound classification process of musical instruments involved 4 main stages, namely pre-processing, feature extraction, feature selection and classification, where the majority of the research focuses on the feature extraction and feature analysis [7,8]. Finding a suitable classifier is vital to improving the classification accuracy and efficiency, while the robust feature set is a major challenge in instrument classification.…”