Environmental Noise Classification using LDA, QDA and ANN Methods

Objective: The impact of feature based environmental noise classification and their efficiencies are discussed in this paper. The aim is to recognize, compare, classifyandidentify with the help of computation and applied techniques. Methods/Statistical Analysis: To perform the task of classification of noise sources, LPC and MFCC were used as input to the classifiers in experimental work. LDA, QDA and ANN are tested for the classification purpose. Findings: Once the source is identified we can address these untoward noisiness class and to minimize their impact to the human perceptions by some means or in speech recognition task to enhance the system recognition efficiency. The source can be identified by analysis of various categories of classifier in association with specific feature of noise source.LDA used with LPC gives an overall efficiency of classification is about 65.1% and with MFCC it is about 77.9%. QDA used with LPC gives an overall efficiency of 72.7% and with MFCC it is about 86.3%. ANN with LPC gives an overall efficiency of classification 83.2% and with MFCC it is about 90%. The MSE’s (mean squared error) of ANN with MFCC are found to be 4.94838-e^-2 (training), 5.33561-e^-2 (validation) and 6.95805-e^-2 (testing) and the %error for the same are 9.40265 (training), 10.02949 (validation) and 14.45427 (testing). Application/improvement: The performance of LDA, QDA and ANN with LPC and MFCC is analyzed. It is evident that ANN in combination with MFCC gives the best result and showing efficiency about 90%.