Classification and Detection of Objectionable Sounds Using Repeated Curve-Like Spectrum Feature

This paper proposes the repeated curve-like spectrum feature in order to classify and detect objectionable sounds. Objectionable sounds in this paper refer to the audio signals generated from sexual moans and screams in various sexual scenes. For reasonable results, we define the audio-based objectionable conceptual model with six categories from which dataset of objectionable classes are constructed. The support vector machine classifier is used for training and classifying dataset. The proposed feature set has accurate rate, precision, and recall at about 96%, 96%, and 90% respectively. With these measured performance, this paper shows that the repeated curve-like spectrum feature proposed in this paper can be a proper feature to detect and classify objectionable multimedia contents.