Vpcid-A Voip Phone Call Identification Database

Audio forensic plays an important role in the field of information security to address disputes related to the authenticity and originality of audio. However, some audio forensics methods presented in existing references were evaluated under either non-forensic oriented databases or private databases which were not publicly available. It creates difficulty for researchers to make comparison between different methods. In this paper we established VPCID, a VoIP phone call identification database for audio forensic purpose. As there is an increasing trend of phone scams or voice phishing via VoIP, through which the caller's identity can be hidden or forged easily, it is demanded to address the issues of identifying VoIP phone calls. The VPCID database is comprising of 1152 VoIP call recordings and 1152 mobile phone call recordings, each of which has more than two minutes. Recordings were collected from 48 different speakers using different smart phones and by considering varies recording conditions such as VoIP software, locations etc. We used MFCC (Mel-Frequency Cepstral Coefficients) and ACV (Amplitude Co-occurrence Vector) based features respectively equipped with SVM (Support Vector Machine) classifier to perform classification on the database. We also evaluated our own database on a CNN (convolutional neural network), but the performance is not too much satisfactory. Therefore the VoIP phone call identification problem is challenging and it calls for more effective solutions to address the problem. We hope our proposed database will convey more than this paper and inspire the future studies, which is openly available in below link, http://media-sec.szu.edu.cn/VPCID. html, and we welcome the use of this database.