Year: 2026
Type: article
Abstract: This paper presents SRLA (Svr-fiR Lossless Audio), a practical lossless audio codec based on linear predictive coding (LPC). SRLA systematically integrates and evaluates three improvements for LPC-based lossless audio coding: (i) Support Vector Regression (SVR)-motivated refinement of LPC coefficients, (ii) adaptive audio-frame segmentation, and (iii) efficient use of higher LPC prediction orders. The study also analyzes a continuous relaxation of the recursive Golomb–Rice (RGR) codeword length under the adopted parameter setting and uses it to motivate an SVR-type coefficient refinement. In the maximum-compression setting, SRLA reduced the compressed-size ratio by 1.04 percentage points and achieved a 2.3-fold lower decoding-time ratio than MPEG-4 ALS, at the cost of a 7.3-fold higher encoding-time ratio. The ablation study further shows that the proposed components have different compression gains and encoder-side costs while having little effect on decoding time.
Authors: Taiyo Mineo, Shouno Hayaru
Institution: University of Electro-Communications
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,190
OpenAlex ID: W7203702057