Multiple Instance Learning for Automatic Content-Based Classification of Speech Audio

Abstract

Speech analytics researchers are working to improve
their ability to decipher audio material. This research presents a
new method for classifying news audio clips based on their
content, called the Multiple Instance Learning (MIL) approach.
Audio classification and segmentation benefit from content-based
analysis. As a starting point, a classifier that can predict the
category of an audio sample has been proposed. Perceptual
Linear Prediction (PLP) coefficients and Mel-Frequency
Cepstral Coefficients (MFCC) are two kinds of features used for
audio content identification (MFCC). For classification, two MIL
approaches, mi-Graph and mi-SVM, are used. Different
performance matrices are used to assess the outcomes gained via
the use of various approaches. The results of the experiments
clearly show that the MIL has great audio categorization
capacity.