Fast software implementation of MPEG advanced audio encoder
I. Dimkoviae, D. Milovanoviae, Z. Bojkoviae · 2003
An optimized software implementation of a high quality MPEG AAC-LC (low complexity) audio encoder is presented in this paper. The standard reference encoder is improved by utilizing several algorithmic optimizations (fast psycho-acoustic model, new tonality estimation, new time domain block switching, optimized quantizer and Huffman coder) and very careful code optimizations for PC CPU architectures with SIMD (single-instruction-multiple-data) instruction set. The psychoacoustic model used the MDCT filterbank for energy estimation and peak detection as a measure of tonality. Block size decision is based on local perceptual entropies as well as LPC analysis of the time signal. Algorithmic optimizations in the quantizer include loop control module modification and optimized Huffman search. Code optimization is based on parallel processing by replacing vector algebra and math junctions with their optimized equivalents with Intel/sup /spl reg// Signal Processing Library (SPL). The implemented codec outperforms consumer MP3 encoders at 30% less bitrate at the same time achieving encoding times several times faster than real-time.