Steganalysis of adaptive multi-rate speech with unknown embedding rates using clustering and ensemble learning

Abstract

With the widespread use of adaptive multi-rate (AMR) speech-based applications, AMR speech-based steganography has witnessed significant growth. Consequently, steganalysis approaches have garnered attention to mitigate network security risks associated with AMR speech-based steganography. However, existing studies often assume known embedding rates of test samples, leaving steganography detection under unknown embedding rates—an encountered practical scenario—unresolved. To tackle this challenge, this paper presents a novel detection scheme for AMR speech-based steganography, skillfully combining clustering and ensemble learning. The training phase utilizes K-means clustering to pre-classify speech samples, grouping them into distinct clusters based on their feature distribution and embedding rates. Subsequently, a classifier based on extreme gradient boosting (XGBoost) is trained for each cluster. The experimental results demonstrate that the proposed scheme exhibits significant improvements in terms of recall rate when compared to existing steganalysis techniques.