The GIR dataset is an open source dataset composed of more than 900k Impulse Responses (IRs). To foster the research in computational acoustics and in particular on the relationship between geometry and acoustic parameters, the measurements include acoustic responses acquired in 2952 positions and 312 surfaces. To maximize consistency, surfaces were 3D printed and the measurement process was automated using two robots. Detailed explanation of the dataset creation methods and possible applications are discussed in Rust et al. 2021 (https://doi.org/10.1177/1351010X20986901) and Xydis et al. 2021 (in prep). The provenance and metadata tracked on Renku is enriched with the additional attributes recommended in the framework dataset nutrition labels and Schema.org note on Data and Datasets. This enriched metadata is provided in a standard JSON-LD format on Renku (dataset.json) and on Zenodo as part of the published dataset (https://doi.org/10.5281/zenodo.5288744). Dataset is released under the GNU General Public License v3.0

To obtain an Impulse Response (IR), we play a linear sine sweep, ranging from 2 to 40 kHz and record the signal with a microphone at 96kHz sampling rate as raw data in an uncompressed WAV file in 32bit resolution. Then, the IR is computed by deconvolution, and by applying temperature compensation (See [Rust 2021, Appendix]). This dataset focuses on the first sound reflexion, so the IR is time-windowed between 0-4 ms. The recording is done for a range of architectural geometries and baseline surfaces. Further post-processing can be applied and a pipeline to do so is available in the code provided with this dataset. We concatenate the IRs of each panel into one '.npz' file, which forms one part of this dataset. Each of the geometries used to print the surfaces is saved in an '.obj' file. Finally, for each panel, we also provide a '.json' file containing fabrication and environmental metadata related to the geometry and the measurement process respectively.