An efficient search method for the content-based identification of telephone-SPAM
Julian Strobl, Bernhard Mainka, Gary Grutzek, Heiko Knospe · 2012
With the help of VoIP technology, large numbers of unsolicited calls can be conveniently placed and SPAM over Internet Telephony may become a major nuisance and threat. Various mitigation methods have been proposed which are mostly based on a pattern analysis of the signaling traffic. This contribution shows that an analysis of the audio content is also feasible and can provide protection against replayed calls. In order to identify similar or equal audio data, spectral features are extracted and a short and robust audio fingerprint is computed. The definition of the fingerprint is optimized for a fast index-based search. Then, the matching of telephone speech data is based on the intersection of inverted files of audio fingerprints. Furthermore, the system design of a working prototype is explained and experimental results on the recognition rate and the performance of the system are presented. It can be shown that the search method is suitable for an efficient identification of SPAM calls.