Evaluating perceptual transparency of lossy music audio compression with Opus codec: dataset and main results

This page provides full experimental details for the reproducibility of the following ICASSP 2027 submission:

Benjamin Martin, Romain Hennequin. "EVALUATING PERCEPTUAL TRANSPARENCY OF LOSSY MUSIC AUDIO COMPRESSION WITH OPUS CODEC".

The structure of the provided files is as follows:

• sample_index.csv provides the list of the 23 audio samples used for the experiment, listing their identifier (number), their source, title, genre and practical Opus encoding bitrates.
• subject_index.csv provides the list of the 26 human subjects involved in the experiment, listing their identifier (letter), gender, age category, musician experience, audio manipulation experience, and self-declared listening expertise.
• results_phase1.csv provides all results gathered in the method's first experimental phase, listing for each trial the subject and sample identifiers, the bitrate tested, the grade assigned to the hidden reference and the grade assigned to the hidden compressed sample.
• results_phase2.csv  provides all results gathered in the method's second experimental phase, with the same field structure as the previous results.

Finally, we supplement the results presented in the paper with per-sample score distributions, organized as follows:

• gradings_per_sample_phase1_part1.png and gradings_per_sample_phase1_part2.png provide the gradings given for each sample in phase 1
• gradings_per_sample_phase2.png provide the gradings given for each sample in phase 2