基于BP-Adaboost强分类器的声音环境识别

Acoustic environment identification based on BP⁃Adaboost strong classifier

针对通过改进分类算法提升声音环境识别正确率的目的,提出了基于BP-Adaboost强分类模型对声音所属环境进行识别的方法。提取声音信号中的梅尔倒谱频系数,通过T分布检验保留高识别度特征,作为声音信号的广义特征;另外,改进传统的分类器训练方式,利用Adaboost算法将多个BP神经网络弱分类器组合成为强分类器,训练声音环境识别模型,根据广义特征对不同环境中的声音进行识别。实验结果表明,BP-Adaboost强分类识别算法相较于传统方法在声音环境识别中的准确率与处理速度上均有显著提升,平均识别率提升了17%,计算效率提升了30%。

For the purpose of improving the accuracy of acoustic environment identification by improving the classification algorithm,this paper proposes a method to identify the environment to which sound belongs based on the BP⁃Adaboost strong classification model.Extract the Mel⁃frequency cepstral coefficients in the sound signal,and retain the high⁃recognition feature through the T distribution test as a generalized feature of the sound signal;in addition,improve the traditional classifier training method and using Adaboost algorithm to combine multiple weak BP neural network classifiers into strong classifiers,and then train an acoustic environment identification model to recognize sounds in different environments based on generalized features.The experimental results show that the BP⁃Adaboost strong classification and recognition algorithm has significantly improved the accuracy and processing speed of acoustic environment identification compared with traditional methods,the average recognition rate has been increased by 17%and the calculation efficiency has been increased by 30%.