ODAQ-Collection

ODAQ-Collection The ODAQ-Collection consolidates the subsets of the Open Dataset of Audio Quality (ODAQ): stereo audio signals processed by a range of methods, together with subjective quality scores (MUSHRA, Basic Audio Quality) from listening tests. Example code for using this collection and running a benchmark experiment is available in the ODAQ GitHub repository. Subsets and Reference Papers 2024-icassp -- M. Torcoli, C.-W. Wu, S. Dick, P. A. Williams, M. M. Halimeh, W. Wolcott, E. A. P. Habets, “ODAQ: Open Dataset of Audio Quality,” ICASSP 2024. (arXiv:2401.00197) https://arxiv.org/abs/2401.00197 2024-aes -- S. Dick, C. Thompson, C.-W. Wu, et al., “Expanding and analyzing ODAQ – the open dataset of audio quality,” AES Convention, 2024. (arXiv:2504.00742) https://arxiv.org/abs/2504.00742 2025-aes -- S. Dick, C. Thompson, C.-W. Wu, P. Delgado, P. A. Williams, M. Torcoli, “Investigating the impact of stereo processing – a study for extending the Open Dataset of Audio Quality (ODAQ),” AES Convention, 2025. (arXiv:2512.14259) https://arxiv.org/abs/2512.14259 2025-lanzendorfer -- L. A. Lanzendörfer, F. Grötschla, “Evaluating objective speech quality metrics for neural audio codecs,” arXiv:2511.19734, 2025. https://arxiv.org/abs/2511.19734 proposed-addition -- C.-W. Wu, M. Torcoli, et al., “ODAQ-Collection: Expanding the Open Dataset of Audio Quality with a Focus on Spatial Distortions,” arXiv:2610.37959, 2026. https://arxiv.org/abs/2610.37959 odaq-collection-content-analysis.csv lists the content type (e.g., speech, music, solo instrument) and the spatial characteristic (e.g., for now mono or stereo) of every unique source item, per subset. Naming Conventions Subset names Subsets are named <year>-<venue>, where the venue is the conference or workshop where the subset was presented. The exception is 2025-lanzendorfer, an arXiv preprint, which is named after its main contributor. proposed-addition is the subset introduced by the ODAQ-Collection paper. Results and listening-test files *_results.csv contain the raw subjective scores. The columns are `score`, `method`, `condition` (Q1–Q6, Ref, LP3.5, Mono), `process` (the file name of the condition), `item`, and `subject`. *.xml are the raw listening-test session files, one per listener. Listening-test folders contain one folder per item. Each item folder holds a reference.wav, the processed conditions, and the anchors (LP35.wav, mono.wav). License files _detailed_license.csv gives the origin, author, and license of each unprocessed source. Processed files carry the license of their source. _license_disclaimer.txt summarizes the licenses and how to cite each package.