Evaluating perceptual transparency of lossy music audio compression with Opus codec

Abstract

Although the open-source Opus audio codec is widely used for digital audio streaming, its perceptual transparency threshold for high-bitrate music compression remains insufficiently evaluated. To establish this threshold, this study evaluates Opus across a two-phase ITU-R BS.1116-3 double-blind framework, analyzing Difference Grades using a Linear Mixed-Effects Model (LMM) and Two One-Sided Tests (TOST) for statistical equivalence. In a first phase with 21 qualified listeners and 23 music excerpts, 128 kbps achieves statistical in-distinguishability ($p = 0.456$) but narrowly fails strict equivalence to the uncompressed reference ($p = 0.064$). In Phase 2, Best Linear Unbiased Predictors (BLUPs) isolate the 10 most critical listeners and 8 most challenging samples for a stress-test. Under these worst-case conditions, 128 kbps is conclusively detected as degraded ($p = 0.030$), whereas 160 kbps successfully reaches perceptual transparency ($p = 0.747$) and strict statistical equivalence ($p = 0.024$). The study concludes that 160 kbps demonstrates perceptual transparency to uncompressed CD quality in critical listening environments. For reproducibility, we publicly release results for both phases, anonymized subject background info and sample characteristics.