Nowadays, Voice over Internet Protocol (VoIP) which enables voice conversation remotely over packet switched networks gains much attentions for its low costs and flexible services. However, VoIP calling anonymity, particularly to withhold "who called whom", is difficult to achieve since VoIP infrastructures are usually deployed in an open networking environment (e.g., the Internet). Our work studies an anonymisation overlay network (AON) based solution to prevent surveillance from external attackers, who are able to wiretap the communication channels as well as to manipulate voice packets in the channels. However, it has been demonstrated that the VoIP combined with traditional AONs are vulnerable to two attacks, namely watermark attack and complementary matching attack. Taking these two attacks into account, we investigate the "defensive dropping" method in VoIP: A VoIP user-agent sends packets to an AON in a constant rate, but packets during periods of silence are marked. Then, the AON drops some silence packets and forwards the remaining ones to their destinations. The result of our experiments shows that the dropping rate must be carefully selected to counteract both of the two attacks. Finally, we discuss further threats in terms of this solution.