关键词: 不良信息过滤, 低俗语音识别, 移位差分倒谱参数, 高斯混合模型, 支持向量机

Abstract: In order to check the spread of objectionable video and sound multimedia information in the internet, this paper proposes an objectionable sound recognition method based on shifted delta cepstral features. The proposed method partitions the input sound signal into frames, and then extracts the shifted delta cepstral features on every frame. The method adopts Gaussian mixture model to preliminary classify the frame. The objectionable frames are verified by Support Vector Machine（SVM）. The experimental results demonstrate the proposed method has high accurate recognition rate and low error recognition rate, and can be used to filter the objectionable sounds and videos in the Internet.

Key words: objectionable information filter, objectionable sounds recognition, shifted delta cepstral features, Gaussian Maxture Model（GMM）, Support Vector Machine（SVM）