The Adaptive Multi-Rate (AMR) audio codec is now widely used as the default file format for various mobile phones to store spoken audio recordings.Meanwhile,more and more people start to use their mobile phones to record sounds for convenience,which results in rapid growth of the amount of digital audio recordings as evidences occur in court,including AMR format audios.It is critical to authenticating the integrity of AMR audios when they appear as evidences.AMR audio forgery manipulations generally uncompressed an AMR file,tamper with the file in PCM domain,and then re-compressed the tampered audio back into AMR format,which come the double AMR compression.In this paper,we propose a method on the detection of double AMR compression by discriminating double-compressed AMR audio recordings from single-compressed ones.Binary classification algorithm is applied for the discrimination.Statistical features related on frequency energy distribution and frequency components correlation are extracted and a support vector machine is applied to execute the classification.Experiment results show our method is promising on this binary classification task.