SoundDesc: Cleaned and Group-Filtered Splits

This upload contains dataset splits of <em>SoundDesc</em> [1] and other supporting material for our paper: <strong>Data leakage in cross-modal retrieval training: A case study </strong>[arXiv] [ieeexplore] In our paper, we demonstrated that a data leakage problem in the previously published splits of SoundDesc leads to overly optimistic retrieval results.<br> Using an off-the-shelf audio fingerprinting software, we identified that the data leakage stems from duplicates in the dataset.<br> We define two new splits for the dataset: a cleaned split to remove the leakage and a group-filtered to avoid other kinds of weak contamination of the test data. SoundDesc is a dataset which was automatically sourced from the BBC Sound Effects web page [2]. The results from our paper can be reproduced using <em>clean_split01</em> and <em>group_filtered_split01</em>. <em>If you use the splits, please cite our work:</em> Benno Weck, Xavier Serra, "Data Leakage in Cross-Modal Retrieval Training: A Case Study," ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Rhodes Island, Greece, 2023, pp. 1-5, doi: 10.1109/ICASSP49357.2023.10094617. <pre><code>@INPROCEEDINGS{10094617, author={Weck, Benno and Serra, Xavier}, booktitle={ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, title={Data Leakage in Cross-Modal Retrieval Training: A Case Study}, year={2023}, volume={}, number={}, pages={1-5}, doi={10.1109/ICASSP49357.2023.10094617}} </code></pre> References: [1] A. S. Koepke, A. -M. Oncescu, J. Henriques, Z. Akata and S. Albanie, "Audio Retrieval with Natural Language Queries: A Benchmark Study," in IEEE Transactions on Multimedia, doi: 10.1109/TMM.2022.3149712. [2] https://sound-effects.bbcrewind.co.uk/