Support vector machines for improved voiceband classification

A method for detecting and classifying the presence of various voiceband data signals on the General Switched Telephone Network (GSTN) is presented. The classification vectors are extracted from processing of the speech parameters evolved by a standard speech coding algorithm. A multi-class Support Vector Machine (SVM) approach is implemented to optimise the classification parameters improving the ability of the system to operate under poor signal-to-noise ratio (SNR) conditions. It is shown that the newly proposed classifier improves on previous implementations and is capable of detecting various 'V' series standards at SNRs well below 12dB.