Accuracy Enhancement of Feature Extraction Scheme in Detection of Chainsaw Sound to Prevent Illegal Logging
This paper proposes an alternative scheme for feature extraction method based on Mel-Frequency Cepstral Coefficients (MFCC) and Sinusoidal Lifter. MFCC method itself works by using a coefficient called mel coefficient to represent features from audio. However, MFCC has had a problem in that the MFCC was susceptible to disturbance of noise. In this proposed scheme, the mel coefficient produced by the MFCC will later be passed through a lifter to solve the noise problem. The result from the experiment conducted on the case of chainsaw sound detection shows the use of lifter has given a higher capability of learning on artificial neuron network, compared to MFCC without lifter as MFCC without lifter requires more than 20000 iterations to be able to minimize error rate to 0.005. The MFCC with lifter, however, has taken only 16 iterations. The MFCC with lifter is, therefore, shown to have produced higher accuracy than MFCC without lifter by 20%.
Paper
Full text
Accuracy Enhancement of Feature Extraction Scheme in Detection of Chainsaw Sound to Prevent Illegal Logging
Semantic Scholar · Engineering · 2019
Abstract
This paper proposes an alternative scheme for feature extraction method based on Mel-Frequency Cepstral Coefficients (MFCC) and Sinusoidal Lifter. MFCC method itself works by using a coefficient called mel coefficient to represent features from audio. However, MFCC has had a problem in that the MFCC was susceptible to disturbance of noise. In this proposed scheme, the mel coefficient produced by the MFCC will later be passed through a lifter to solve the noise problem. The result from the experiment conducted on the case of chainsaw sound detection shows the use of lifter has given a higher capability of learning on artificial neuron network, compared to MFCC without lifter as MFCC without lifter requires more than 20000 iterations to be able to minimize error rate to 0.005. The MFCC with lifter, however, has taken only 16 iterations. The MFCC with lifter is, therefore, shown to have produced higher accuracy than MFCC without lifter by 20%.