Research Of Analysis And Characteristics Extraction On Fish Sound

Fish is very rich in our country. So it is important to classify the fish using the acoustic technology. This paper analyzes the sound of fish using cepstrum and Hilbert-Huang transform methods, and then extracts the characteristics.The Hilbert-Huang transform (HHT) is a new theory, which is first developed by Huang at the end of last century for the signal analysis. This paper disserts the idea and how to achieve of HHT in detail. This paper also analyzes the frequency resolution by simulation, points out the weakness of the classic HHT method, and develops the HHT algorithm using the wavelet packet predecomposition and normalized correlation determination. The frequency resolution is improved.Fish vocalizes in its own way. This paper introduces three ways of fish making sound and the two methods used in analyzing the sound of fish. The two methods is time domain analysis and cepstrum analysis. Time domain analysis can get the voice sound. The cepstrum analysis can get the characteristics.Characteristics extraction is important in analyzing fish sound. This paper gets two types of characteristics. One of them is Mel-frequency cepstral coefficients(MFCC) and its first difference. Another is the Hilbert bound spectrum. Before getting the Hilbert bound spectrum pre-weighting the fish sound will be easy to get the vocal tract characteristics. At last, this paper gets the Mel-frequency cepstral coefficients and the Hilbert bound spectrum, classifies five different fish species using BP artificial neural network, then discusses whether the experimental results fit with the theoretical discussion or not. The results show that these two methods can get good results of classification.With the result of simulation and actual data, both the Mel-frequency cepstral coefficients and the Hilbert bound spectrum can be used in the classification of fish. Using these characteristic both can get good results. So the HHT and cepstrum analysis methods are effective methods of fish sound characteristics extracting.