Inpainting techniques are becoming increasingly important for lossy image compression. In this paper, we investigate if successful ideas from inpainting-based codecs for images can be transferred to lossy audio compression. To this end, we propose a framework that creates a sparse representation of the audio signal directly in the sample-domain. We select samples with a greedy sparsification approach and store this optimised data with entropy coding. Decoding restores the missing samples with well-known 1-D interpolation techniques. Our evaluation on music pieces in a stereo format suggests that the lossy compression of our proof-of-concept framework is quantitatively competitive to transform-based audio codecs such as mp3, AAC, and Vorbis.
CITATION STYLE
Peter, P., Contelly, J., & Weickert, J. (2019). Compressing Audio Signals with Inpainting-Based Sparsification. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11603 LNCS, pp. 92–103). Springer Verlag. https://doi.org/10.1007/978-3-030-22368-7_8
Mendeley helps you to discover research relevant for your work.