Classification of Broadcast Audio using Random Forests and SVM with MFCC Features

B. Kamatchy and Dr.P. Dhanalakshmi

Television has become a part of our day by day life. There is a developing assortment of enthusiasm for classifying TV broadcast programs. In this paper we proposed the technique MFCC-Mel Frequency Cepstral Coefficient for extracting acoustic features from TV broadcast audio and classify the audio into one of the five predefined classes such as Advertisement, News, Cartoon, Songs and Sports using efficient algorithms. The Cepstral Coefficients of Mel Frequency [MFCCs] are utilized widely for audio classification and recognition. MFCCs are a small set of features that typically define the overall shape of a spectral envelope in a concise manner usually around 10-20 ms. The extracted features from each class are given to the machine learning techniques namely Random Forests and SVM for classifying TV broadcast data. Data set is collected from television using tv tuner card. Performance of the models Random Forests and SVM is also compared in the proposed model.

Volume 12 | 03-Special Issue

Pages: 181-187

DOI: 10.5373/JARDCS/V12SP3/20201252