# ClapRIR website assets

Three groups: paper-linked examples, website-only extensions, and dataset recording material. The bundle reuses existing inputs and model estimates; no inference or training was run.

## Paper-linked material

The two main rooms use exactly the frozen paper membership: one unclipped recording for each of eight modes, chosen by highest pre-handclap SNR before inference. The meeting-room predictions come from the authoritative saved figure arrays, not substituted full-corpus cache rows. Raw arrays and per-asset hashes are included. The six-pair Slack bundle is at ../audio_examples.zip.

## Website extensions

All 280 normal phone protocol events (7 rooms x 8 modes x 5 repetitions) have paired recorded/inferred listening WAVs. The 56 preselected demo events feed the room spectra and spectrograms. All 13 existing room photographs have documented room identities and are copied unchanged. Room labels and exported filenames use Highly reverberant hallway; original source paths preserve original filenames for traceability.

## Processing

Audio: mono 44.1 kHz IEEE float WAV. Phone files preserve the complete one-second model interval and original alignment, including existing input padding where valid recording support is shorter. Each listening file alone is peak-scaled to -1 dBFS. No extra filtering, denoising, onset shifting or inference is applied. Every export is read back and checked for exact scaled sample equality, rate, duration, finite samples and no clipping. Original peaks/gains are in manifest.csv.

Website spectra: 0.1--10 kHz default, common 0--0.9-s interval, 1/12-octave mean linear FFT power, per-signal mean FFT-bin power normalization over 0.1--10 kHz. Original 15-kHz paper figures and a separately labeled existing 20-kHz diagnostic are additional assets. Spectrograms: Hann 1024, hop 128, FFT 1024, boundary zeros, no padded final STFT frame; 0--0.9 s and 0.1--10 kHz; normalization by each panel mean power; shared -20 to +30 dB. Waveforms: 0--250 ms. Only the temporal panels of the two current combined paper figures remove per-signal normalization. The spectral panels retain the original mean FFT-bin power normalization over 0.1--10 kHz and 1/12-octave power averaging, displayed over 0.1--15 kHz with shared -40..20 dB limits. RMS panels use original-sample RMS (221-sample centered window, 44-sample spacing) and 20 log10(RMS) with common digital full scale one, shared -60..0 dBFS limits. The linear alternative preserves original signed samples, shared -1..1 limits. No temporal peak/RMS-maximum division or fitted gain is applied. Temporal levels are digital, not calibrated SPL. Archived comparison_peak_normalized figures and the separate P1/P2/P3 strip retain their previous normalization. All source signals are unchanged.

## Dataset material and release status

Five anechoic listening examples use the same source events as fig_clap_dataset, from participants 01, 02, 04, 05 and 07, preserving their complete retained source clip rather than the figure 20-ms crop. Individual-event clap-mode labels are unknown and are left blank. The local legacy event inventory and the newer 2540-event release description in manuscript edits are distinct; no newly released dataset statistics or mode assignments are inferred from the legacy files.

The phone audio, verified room photographs, generated figures and unlabelled anechoic examples are prepared for the user-requested public demo. The repository software license is recorded separately. The supplied clap_spectra_with_hands PDF is preserved exactly, with a raster preview. No event-level pose assignments, setup photograph, missing paper/preprint link, or dataset license has been invented. No paired phone reference RIR exists, and no reconstruction-accuracy claim is made. Perceptual plausibility remains a human listening judgment.

## Missing or not yet suitable for publication

- Anechoic recording setup image clap_measurement.jpeg is referenced in the manuscript but absent from the local publication/data material.
- Separate pose-image files and event-level pose mapping for the five legacy anechoic examples remain unverified. The user-supplied clap_spectra_with_hands composite now provides the labelled protocol illustration; it does not label those separate listening events.
- Public full anechoic-dataset download URL not present in the inspected sources.
- Public paper/preprint URL and downloadable model checkpoint link not present in the inspected sources.
- Separate reuse/license terms for dataset audio and room photographs were not present; the software MIT license is not extended to them.

Source identifiers, checkpoint SHA-256, original peaks, gains, selected-for-paper flags and exact export paths are recorded in manifest.csv, phone_demo/all_rooms/recording_inventory.csv, and the JSON inventories. The manuscript was not modified. The combined figures preserve original temporal levels and retain the original mean-power spectral normalization; source signals and predictions are unchanged.