Journal of Digital Information Management Jing XU, Jeffrey XIA 13 2 2015 Digital audio detection is a forensics authenticity request. However, the classic methods for digital audio detection are no longer effective, especially given the use of compound tampering, such as adding background noise, and the size of processed data sets easily exceeding gigabytes. This article aims to discuss a detection method that uses a sparse representation classifier based on adaptive least squares (recursive least squares sparse representation classification [RLS-SRC]) and periodicity in the second derivative of an audio signal as a classification feature for digital media forensics. Using adaptive least squares, the proposed RLS-SRC can perform online updates and thus reduce the burden of training. In cases with background noise, our proposed method yields better classification compared with the method based on K-singular value decomposition (KSVD).