Mel sub-band parameterized feature-based warble automatic recognition method

Abstract

The invention discloses a Mel sub-band parameterized feature-based warble automatic recognition method. According to the method of the invention, as for a warble-based species taxonomy problem under the condition of a natural complex acoustic environment, a Gaussian mixture model is used to fit the logarithmic energy distribution of framed continuous acoustic monitoring data; data frames with highlikelihood ratios are selected to form candidate sound events; robust detection and automatic segmentation are completed on the basis of a candidate event energy-based post-processing process; a Melband-pass filter bank is adopted to perform filtering processing on corresponding segments in a spectral domain, the energy sequences of various sub-band output which varies with time are modeled based on an autoregressive model, so that parameterized features which can describe the time-frequency features of different kinds of warble signals can be obtained; and a support vector machine algorithmis used for performing classification and recognition. The method of the invention is simple to implement and has important significance for ecological monitoring.

Description

Technical field [0001] The invention belongs to the technical field of ecological monitoring and acoustic signal recognition, and specifically is an automatic recognition method of bird song based on the parameterized feature of the Mel subband. Background technique [0002] As an important part of ecological monitoring, biodiversity monitoring has very important significance in studying species resources and ecological environment. Among them, birds, as a kind of environmental indicator organisms that are extremely sensitive to habitat changes and environmental changes, are important indicators for biodiversity assessment and monitoring, and ecological environmental impact assessment. Automatic bird monitoring based on bird song is a non-invasive method of detecting the existence and activity of birds, which is of great significance to ecological research. [0003] Among them, the use of acoustic signal analysis methods to extract features of bird sound signals measured in the fi...

AI Technical Summary

Problems solved by technology

In recent years, some scholars have proposed to extract the texture feature vector corresponding to the spectrum of bird song fragments based on the sum-difference statistics method (Wei Jingming, Li Ying. Fast bird song recognition using anti-noise texture features[J]. Electronic Journal, 2015, 43( 1): 185-190.), but the data samples used in the experimental verification are all manually intercepted segments with a fixed length of 2 seconds, which do not involve the automatic detection and extraction of vocal segments. application, the method of manually intercepting segments is not feasible, and the above method has poor performance in classifying vocal segments with varying durations

[0004] It can be seen that there are defects such as incomplete methods and single functions in the prior art.

Images