Lossless audio codec based on CNN, weighted tree and arithmetic encoding (LACCWA)

In this paper, an integrated lossless audio codec with improved compression ratio has been proposed using traditional methods as well as convolutional neural network (CNN) architecture without losing any audio information. The method applies adaptive arithmetic encoding followed by binary weighted tree based transform and convolutional neural network (CNN) respectively. The first level of compression uses adaptive arithmetic encoding to process audio samples block by block, reducing the size of the data to the number of blocks. Subsequently, each of the arithmetic encoded values for every block is encoded by the binary weighted tree based encoding. Estimating a dynamic weighted path, it transforms the arithmetic-encoded data into an equivalent binary pattern. The binary stream is further compressed with latent space representations using a proposed CNN architecture. The analysis of the simulation results are performed using various statistical and robustness characteristics and comparison with other existing methods.