Portal de Periódicos da CAPES

A Method Based on CNN + SVM for Classifying Abnormal Audio Indoors

2021; Springer Nature; Linguagem: Inglês

10.1007/978-981-33-4575-1_37

2194-5357

Jian Liu, Shuyan Ning, Sanmu Wang, Jiarui Yi, Zhao Mingrui,

Speech Recognition and Synthesis

In[aut]Liu, Jian this[aut]Ning, Shuyan paper[aut]Wang, Sanmu, a[aut]Yi, Jiarui novel[aut]Zhao, Mingrui classification algorithm combined convolutional neural network with support vector machine (CNN + SVM) is proposed for classifying abnormal audio indoors to solve the emerging problems, for which video surveillance may have obstacles, blind spots and therefore cannot protect the privacy under the scenarios. First of all, in the experiments, the quality of the audio signal is improved by pre-emphasis, framing, and windowing methods. Secondly, to obtain sufficient audio information, Mel frequency cepstral coefficient is selected as a parameter for feature extraction. Lastly, multilayer perceptron (MLP), convolutional neural network (CNN), support vector machine (SVM), and CNN + SVM are used to classify eight types of audio signals according to the complexities of the indoor environment. The result of the proposed experiments indicates that the CNN + SVM combination algorithm exhibits a higher accuracy rate for the classification of audio compared to that of the traditional single classification algorithm. It outperforms other methods for indoor abnormal audio classification in terms of applicability.