During my work with Google's AudioSet I encountered some problems due to the fact that Weak [1] and Strong [2] versions of the dataset use different csv formatting for the data, and that also labels used in the two datasets are different and also presented in files with different formatting.

This repository aims to unify the formats of the datasets so that it is possible to analyse them in the same pipelines, and also make the dataset files compatible with psds_eval, dcase_util and sed_eval.

The source folder contains also the original dataset files and the script convert.py to regenerate the reformatted files. It can be run from this folder as python src/convert.py, it does not require any external libraries.

NOTE: audioset_weak_train_unbalanced.tsv has been manually split into two files to comply with GitHub's size limit. The Zenodo version has only one big file.

All files are converted to tab-separated *.tsv files (i.e. csv files with \t as a separator). All files have a header as the first line.

Fields are renamed according to the following table, to be compatible with psds_eval:

For class label files, id is now the name for the for mid label (e.g. /m/09xor) and label for the human-readable label (e.g. Speech). Index of label indicated for Weak dataset labels (index field in class_labels_indices.csv) is not used.

Files are renamed according to the following table to ensure consisted naming of the form audioset_[weak|strong]_[train|eval]_[balanced|unbalanced|posneg]*.tsv:

Only changes to the Strong dataset are renaming of fields and reordering of columns, so that both Weak and Strong version have filename and event_label as first two columns.

• Labels are given one per line, instead of comma-separated and quoted list

• To make sure that filename format is the same as in Strong verson, the following format change is made:

The value of the start_seconds field is converted to milliseconds and appended to the filename with an underscore. Since all files in the dataset are assumed to be 10 seconds long, this unifies the format of filename with the Strong version and makes end_seconds also redundant.

Class labels from both datasets are merged into one file and given in alphabetical order of ids. Since same ids are present in both datasets, but sometimes with different human-readable labels, labels from Strong dataset overwrite those from Weak. It is possible to regenerate class_labels.tsv while giving priority to the Weak version of labels by calling convert_labels(False) from src/convert.py.

Both the original dataset and this reworked version are licensed under CC BY 4.0

NOTE Labels come from the AudioSet ontology, which is licensed under CC BY-SA 4.0, but is not replicated in this repository.

This dataset is also available on Zenodo (https://zenodo.org/record/7096702). Please cite that as Bakhtin. (2022). Google's Audioset: Reformatted (v1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7096702.

[1] Gemmeke, J. F., Ellis, D. P., Freedman, D., Jansen, A., Lawrence, W., Moore, R. C., ... & Ritter, M. (2017, March). Audio set: An ontology and human-labeled dataset for audio events. In 2017 IEEE international conference on acoustics, speech and signal processing (ICASSP) (pp. 776-780). IEEE.

[2] Hershey, S., Ellis, D. P., Fonseca, E., Jansen, A., Liu, C., Moore, R. C., & Plakal, M. (2021, June). The benefit of temporally-strong labels in audio event classification. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 366-370). IEEE.