Real measurement audio — functional oracles with published values#

Third-party measurement recordings and series derived from them, included in this repository for validation only: each file comes with a value its authors published, and the test suite computes that value through the library's own reading and measurement chain. They are test data, not part of the phonometry package; they are not installed with it, and this repository's MIT licence does not cover them — the licences named below do. The general rules for committed oracle data are in tests/data/README.md.

What this folder contains#

Source, licence and attribution, file by file#

xl2/ — NTi XL2 pass-by campaign, University of Antwerp (CC BY 4.0)#

Attribution: "Calibration file and pass-by excerpts from: A. Grangeiro de Barros and C. Vuye, 'Psychoacoustic indicators of pass-by road traffic noise', Zenodo, 2023. DOI 10.5281/zenodo.7904680. Licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Recorded with an NTi Audio XL2 sound level meter." Retrieved from the Zenodo deposit on 2026-08-18 (archive checksum verified against the record).

• calibration_113_7dB.wav is byte-identical to Calibration file 1Hz 113.7 dB.wav of the deposit (renamed here; the "1Hz" of the original name is the depositors' typo for the 1 kHz calibrator tone). PCM 24-bit mono 48 kHz, written by the meter itself: the bext chunk carries Originator "NTi Audio XL2 A2A-17367-E0" and Description "0dBFS = 129.3 dBSPL". Published value: the deposit names and describes it as the 113.7 dB calibration recording of the campaign's sound level meter and microphone.
• passby_*.wav are five 4-second pass-by excerpts (one per vehicle category and speed named in the file name), cut bit-exactly (whole 24-bit sample codes, unmodified) from the campaign's hour-long session recordings at the sample ranges given by the deposit's timestamp workbooks, and repackaged in a minimal fmt +data RIFF container. The session bext chunks are deliberately not copied into the cuts, so no instrument metadata is misattributed to an excerpt the instrument did not write; each session's digital full scale (0 dBFS = 129.4 or 129.5 dB SPL) is transcribed in the consuming test instead, next to the published values it calibrates. Changes with respect to the original files: cut and recontainered as described; sample data unmodified. Published values: the deposit's master workbook Psychoacoustic indicators of pass-by road traffic noise.xlsx lists, per pass-by, LA,eq, LA,max, L10 and the median Zwicker loudness N50 computed by the authors; the five rows used (indices 566, 571, 668, 960 and 1558) are transcribed in the consuming test.

sounddevices/ — wavinfo test file (MIT)#

Attribution: "A101_3.WAV from the wavinfo project (https://github.com/iluvcapra/wavinfo), MIT License, Copyright (c) 2018-2023 Jamie Hardt." Retrieved from the wavinfo repository on 2026-08-18.

A real BWF written by a Sound Devices 702T field recorder: bext version 1 with every string field populated by the machine, plus a 5.2 kB iXML chunk. Byte-identical to tests/test_files/sounddevices/A101_3.WAV of the wavinfo repository. The wavinfo documentation documents this recorder's bext output on the sibling take A101_1.WAV; the expected field values for this file are transcribed in the consuming test and were cross-read with an independent reader (ffprobe) at import time.

libbw64/ — EBU libbw64 test file (Apache License 2.0)#

Attribution: "rect_24bit_rf64.wav from the EBU libbw64 project (https://github.com/ebu/libbw64), Apache License 2.0 (https://www.apache.org/licenses/LICENSE-2.0)." Retrieved from the libbw64 repository on 2026-08-18.

A genuine RF64 file (RF64 FourCC with a ds64 chunk), PCM 24-bit 44.1 kHz stereo, byte-identical to tests/test_data/rect_24bit_rf64.wav of the libbw64 repository. It is the small real-world RF64 the reader is validated against; the expected sample codes of its rectangular wave are asserted exactly.

nikolauskapelle/ — St. Nicholas Chapel RIR, FH Aachen (CC BY 4.0)#

Attribution: "Omnidirectional channel of the S1-Do_R1 impulse response from: M. Zerwas and S. Kayku, 'Room acoustic measurement and simulation data of the St. Nicholas Chapel, Aachen Cathedral', Zenodo, 2026. DOI 10.5281/zenodo.20428705. Licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)." Retrieved from the Zenodo deposit on 2026-08-18.

s1_do_r1_w.npz holds the W (omnidirectional) channel of the B-format impulse response S1-Do_R1.wav from the deposit's impulse_responses.zip, as w (float32 samples exactly as stored in the original file) and fs (48000). Changes with respect to the original file: one channel of four extracted and repackaged as npz; sample data unmodified. Published values: the deposit's workbook Nikolauskapelle_Source1_Results_20260205.xlsx tabulates the ISO 3382-1 room acoustic parameters (EDT, T20, T30, C80, C50, D50, Ts) of this source-receiver pair per one-third-octave band; the rows used are transcribed in the consuming test.

dublin/ — DCC Ambient Sound Monitoring Network (CC BY 4.0)#

Attribution: "Five-minute LAeq series from the Dublin City Council Ambient Sound Monitoring Network (Ballymun and Ringsend monitors, 2015), Smart Dublin / data.gov.ie open data, dataset 'Ambient Sound Monitoring Network'. Licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)." Retrieved from data.gov.ie on 2026-08-18, together with the annual report from dublincity.ie.

dcc_laeq_5min_2015.npz is a derived series, not audio: for each site, <site>_time0_s (first interval-end stamp, in seconds since 2015-01-01 00:00 local time), <site>_dt_s (differences between consecutive stamps) and <site>_laeq_cdb (the five-minute A-weighted Leq values in hundredths of a dB, an exact transcription of the two-decimal source). Changes with respect to the original files: the daily text logs were concatenated, sorted, de-duplicated (49 stamps re-measured around a monitor restart on 2015-05-15 keep their first value) and repacked as npz; the level values are unmodified. Published values: the DCC "Ambient Sound Monitoring Network Annual Report For 2015" (dublincity.ie), p. 11, gives the annual night-time (23:00-07:00) levels — Ballymun 58 dB(A), Ringsend 59 dB(A) — with logarithmic (energetic) averaging declared.

Purpose and scope of use#

These files exist so that the reading chain (phonometry.io) and the measurement functions can be validated end to end against real recordings with independently published values, in CI, without any local download. Every licence above permits redistribution: two deposits and the Dublin series under CC BY 4.0 (attribution given here, changes indicated above), one file under MIT and one under Apache License 2.0 (notices preserved here). What is deliberately not committed: the campaigns' full recordings (13.2 GB of pass-by audio, the 271 MB impulse-response set), the deposits' workbooks, and the four-channel version of the RIR — the tests transcribe the handful of published numbers they assert instead.

Removal policy#

If you are an author or rights holder of any of these datasets and consider that this redistribution exceeds the intended use, please open an issue or contact the maintainer (see CITATION.cff) and the file will be removed promptly. Each file is consumed by exactly one test module (table above), which would be removed with it; nothing else in the repository depends on this folder.