Abstract: MP3 speech tampering process usually includes three operations, i.e., decompression, tampering, recompression, which will inevitably introduce a double compression. Therefore, double compression detection is a necessary condition for MP3 speech tampering detection. In this work, the statistical characteristics of quantized MDCT coefficients are analyzed, and a model is established for fitting the relationship between speech duration and those characteristics. Finally, a method is put forward, which can not only effectively detect double compression for MP3 speech, but also reveal the relationship between speech duration and detection accuracy.

Key words: double compression, MP3 speech, Modified Discrete Cosine Transform（MDCT） coefficients, fitting

摘要： 通常MP3语音的篡改过程会经过解压、篡改、重压缩三个操作，它不可避免地会产生双压缩。因此MP3语音的双压缩检测是MP3语音篡改检测的必要条件。通过分析量化MDCT系数在不同时长下的统计特征，建立拟合模型，提出一种与时长相关的相同码率MP3双压缩检测方法。该方法不仅可以快速有效地实现对MP3语音的双压缩检测，而且还揭示了语音时长与检测率的关系。

关键词: 双压缩, MP3语音, 改进的离散余弦变换（MDCT）系数, 拟合