随机森林和大规模声学特征的噪声环境鸟声识别仿真

Bird Sound Classification Simulation in Noisy Environment Based on Random Forests and Large Scale Acoustic Features

介绍了一种基于随机森林算法和大规模声学特征的噪声环境下鸟声识别方法。实验基于由德国柏林自然科学博物馆提供的真实鸟声数据以及人工加入信噪比依次为-10 d B、-5 d B、0 d B、5 d B和10 d B的2种类型噪声(即真实环境的背景噪声和高斯白噪声),对60类亚种鸟声进行大规模声学特征提取并进行基于随机森林算法的机器学习。结果表明:该方法对2类噪声环境均具有良好的鲁棒性,并能在较低信噪比时仍具有较好的识别性能。

The paper presents a bird sound classification method based on random forests and large scale acoustic features in noisy environment. The sound data which include 60 sub-classes of birds were provided by Museum for Nature in Berlin,Germany. Two kinds of noises were manually added into the data at the level of-10 dB,-5 dB,0 dB,5 dB and 10 dB signal-to-noise-ratios（SNRs）,respectively（real-world background noises and Gaussian white noises）. Experimental results prove that,the proposed method has excellent robustness in noisy environment and retains good recognition performance at low SNR conditions.