Seal call recognition based on general regression neural network using Mel-frequency cepstrum coefficient features

Abstract In this paper, general regression neural network (GRNN) with the input feature of Mel-frequency cepstrum coefficient (MFCC) is employed to automatically recognize calls leopard, ross, and weddell seals widely overlapping living areas. As a feedforward network, GRNN has only one parameter, i.e., spread factor. The recognition performance can be greatly improved by determining factor based on cross-validation method. This paper selects audio data above three kinds compares machine learning models for inputting MFCC features low-frequency analyzer recorder (LOFAR) spectrum. results show that at same signal-to-noise ratio (SNR), result better than LOFAR spectrum, which verified statistical histogram. Compared other models, still achieve effective low SNRs. Specifically, accuracy 97.36%, 93.44%, 92.00% 88.38% cases an infinite SNR 10, 5 0 dB, respectively. particular, least training testing time. Therefore, all proposed method excellent seal call recognition.