Classification of speech transmission channels: Landline, GSM and VoIP networks

In this paper, we study the classification of three speech transmission channels: landline telephone, mobile phone and voice over Internet protocol (VoIP), based on speech signals collected from these channels. The problem is formulated as a three-class statistical pattern classification problem. The Mel-frequency cepstral coefficients (MFCC) are used as the features for classification and the Gaussian mixture model (GMM) is used to model the distribution of the features. The maximum likelihood (ML) is used as the decision rule for the classification. Our major contribution is that we use different databases for training and testing, so the evaluation tests are completely open. In such tests, high classification accuracy around 95% is obtained which indicates that the classification of speech transmission channels using training data is possible. We also consider factors that may influence the performance of the classification, such as the length of speech used to make a classification decision and the complexity of the GMM.