Comparative Study on Short Time-Frequency and Time Domains for Frog Identification System

Automatic frog sound identification system is one of the useful approaches to assist experts in identifying frog species and to replace manual techniques claimed to be costly and time consuming. However, to execute an automatic system in noisy environment due to background noises is a challenging task. Instead of depending on physical observation procedure to identify the particular species, this study proposes an automated frog identification system based on bioacoustics signal analysis. Experimental studies of 15 species of frogs are used in this study. These calls are then corrupted by 20dB 10dB and 5dB in different stationary and nonstationary noises. The calls are segmented with three different techniques which are Sinusoidal Modeling (SM), combination of Short Time Energy (STE) and Short Time Average Zero Crossing Rate (STAZCR) (STE+STAZCR) and combination of Energy (E) and Zero Crossing Rate (ZCR) (E+ZCR). A syllable feature extraction method i.e. mel-frequency cepstrum coefficients (MFCC) employed to extract the segmented signal. Subsequently, k nearest neighbor (kNN) are employed in order to evaluate the performance of the identification system. Two experiments have been experimented to compare the performace of SM, E+ZCR and STE+ZTAZCR. The classification performance for three techniques are found to be 90.330/0, 93.340/0 and 93.21% for the SM, E+ZCR and STE+ZTAZCR, respectively.