An improved QIM (Quantization Index Modulation) steganography was proposed and it can be used for efficient information hiding in G.729A compressed speech stream.This paper wants to detect this type of steganography.This paper proves that such steganography will significantly change the imbalance and correlation distribution characteristics of codeword (quantization index) in the stream.And then it designs statistical models to extract the quantitative feature vectors of these characteristics.Combining the extracted vectors with the support vector machine

this paper constructs an ensemble classifier for detecting the QIM-based steganography in the G.729A speech stream.The experimental results show that the classifier can achieve up to 98% correct detection towards G.729A encoded speech stream with detection time less than 30 milliseconds.