We present a new algorithm for improved prediction of nonstationary frames for asymmetrical lossless audio compression. Linear prediction is very efficient for decorrelation of audio samples, however it requires segmentation of the audio into quasi-stationary frames. Adaptive segmentation tries to minimize the total compressed size, including the quantized prediction coefficients for each frame, thus longer frames which are not quite stationary may be selected. The new algorithm for computing the linear prediction coefficients improves compressibility of nonstationary frames when compared with the least squares method. With adaptive segmentation, the proposed algorithm leads to small but consistent compression improvements up to 0.56%, on average 0.11%. For faster encoding using fixed size frames, without including adaptive segmentation, it significantly reduces the penalty on compression with more than 0.21% on average.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
