jmrp.io / PyOctaveBand

Fork 0

[READ-ONLY] Mirror of https://github.com/jmrplens/PyOctaveBand. [Python3] Octave-Band and Fractional Octave-Band filter. For signal in time domain. jmrplens.github.io/PyOctaveBand/

acoustics audio filter frequency frequency-analysis frequency-domain octave python3 signal time-domain

Fork 0

PyOctaveBand / tests / data / broadcast /

at main 2 files

José M. Requena Plens Run the STIPA and EBU programme oracles in CI from committed data (#443) 2mo ago

14fa426f

README.md

Run the STIPA and EBU programme oracles in CI from committed data (#443) * Run the STIPA and EBU programme oracles in CI from committed data The certified IEC 60268-16 STIPA bench from stipa.info and the EBU loudness test set only ever ran on a machine holding the private download, so two suites that verify normative criteria were green everywhere by skipping. The STIPA bench is 133 MB of 48 kHz PCM that does not compress as audio, and Embedded Acoustics publishes it for free commercial and non-commercial use, so tests/data/stipa/ now carries a lossless extract of 27 of its 49 signals in 5.8 MB: each signal is stored as an LZMA-compressed difference series with the SHA-256 of its original samples, verified on every load, so what the tests analyse is bit-exact with what stipa.info publishes. The selection keeps the C.3.3 and C.4.2 suites complete and the worst-case point of the other three (C.3.2 m = 0,9 with the m = 0 and m = 1 ends, A.2.2 the outer band pairs, A.3.1.2 TI = 0,9), so the filter-slope, phase-distortion, modulation-depth and weighting-factor criteria are all checked in CI with no tolerance relaxed. The intermediate points of the three partial suites still run wherever the full bench is available. The EBU set is the opposite case: its terms of use forbid copying and distribution outright, and cases 7/8 and 5/6 are licensed film excerpts on top of that, so no file of it can be committed. test_ebu_material_oracle.py therefore takes the committed block-loudness series as its primary path and asserts the published -23.0 LUFS and 5/15 LU targets everywhere, folding in the former test_ebu_material_series.py. Where a local copy exists it still runs the full chain from the audio, and it now re-measures the committed envelopes and checks them against the file, so the derived oracle cannot drift away from the material it stands for. Non-redistributable material also moves out of plan/ into the gitignored tests/data-local/, and every suite that uses it resolves the same way through the new tests/oracle_data.py: environment override, then tests/data-local/, then the committed data - the path CI takes, which never skips. pytest prints which copy each dataset resolved to in its run header, so a green run says which oracle produced it. tests/data/README.md writes down the convention, the licensing evidence per set and what each committed oracle can and cannot assert. * Pin the extract selection by name and make the archive reproducible Counting signals per suite let a same-suite substitution through: swapping the C.3.2 m = 0,9 worst case for an easier point would have kept the count right and quietly weakened the committed coverage. The inventory guard now asserts the exact 27 names. The generator also rejects a source file that is not 48 kHz instead of letting the wrong rate surface later at decode time, and stamps a fixed timestamp on every archive member so rebuilding the extract is byte-reproducible. The STIPA data README now states exactly what the stipa.info free-of-charge grant does and does not say about redistribution, rather than leaving the reader to infer it. * Cite the archived stipa.info page as the record of the licence terms The live download page is currently defaced: it serves an injected script that rewrites the document. The licence text underneath is intact and unchanged, but a licence claim in a public repository should not rest on a page an attacker has had write access to, so the Internet Archive capture of 2025-09-10 is now the citation of record and the live URL is kept as secondary. The Archive holds captures from 2022-04-23 to 2025-09-10, all predating the compromise, and the terms are identical in the first and the last of them. The same capture also sharpens the argument. The page conditions redistribution explicitly where its author means to, on the manufacturers' measurement signals, and attaches no such condition to the verification signals, so the reading this extract rests on is a contrast in the text rather than merely an absence of prohibition. That reading is now stated as ours rather than as a permission granted to this repository, and the removal policy remains the remedy. * Treat a partial local set as absent, and rebuild the extract atomically Three fixes from review. A local copy was accepted as the full set on the strength of one marker path. The STIPA bench ships as five separate downloads, so having unpacked one of them was enough to make the harness enable all five suites and then look for files that were never downloaded, failing with FileNotFoundError instead of falling back to the committed extract. Every dataset now lists all the paths that must be present, and an incomplete copy resolves to the committed oracle. A file missing from a copy that did pass the check reports what is missing and where, rather than substituting the extract behind the reader's back. The extract generator wrote straight into the committed archive, so a missing or malformed source file part way through destroyed the previous good copy. It now builds beside the target and moves it into place only after the archive closes cleanly. The documentation claimed the CI path never skips. The assertions on committed data never skip, which is the property worth having; six EBU cases that read programme audio and the full-bench inventory guard do skip there by design. Said that way in all six places it was written.

2 months ago

ebu_programme_block_loudness.npz

feat: programme loudness and true peak (ITU-R BS.1770-5 / EBU R 128) (#219) * refactor: extract the shared inter-sample-peak oversampling helper The polyphase oversample-then-max machinery of lc_peak also underpins the BS.1770-5 true-peak meter, so it moves to _internal/peaks.py as inter_sample_peak and lc_peak delegates to it (bit-identical behaviour). * feat: programme loudness and true peak (ITU-R BS.1770-5 / EBU R 128) New broadcast domain implementing the objective multichannel loudness measurement algorithm and its EBU companions: - k_weighting / k_weighting_coefficients: the Annex 1 two-stage pre-filter with the Table 1-2 biquads verbatim at 48 kHz and an analog-prototype redesign at other rates (response preserved within 0.016 dB). - integrated_loudness: 400 ms gating blocks at 75 % overlap, Table 3 channel weights (LFE excluded) and the two-stage -70 LKFS / -10 LU gate (Formulae 1-7). - program_loudness: the full EBU Mode set of Tech 3341 as a frozen ProgramLoudnessResult with .plot() (momentary and short-term series with maxima, integrated loudness, relative threshold, loudness range with its percentile edges, per-channel true peaks). - loudness_range: the Tech 3342 algorithm with its cascaded -70 LUFS / -20 LU gate and 10th-95th percentile spread, following the reference implementation's nearest-rank indexing. - true_peak_level: the Annex 2 inter-sample peak in dBTP by oversampling to at least 192 kHz (4x at 48 kHz), sharing the lc_peak machinery. - channel_weight: the Annex 3 position-dependent weights (Table 4) for the BS.2051 advanced sound systems. * test: EBU Tech 3341/3342 minimum-requirements signals for programme loudness Every synthesizable case runs at its official tolerance: Tech 3341 cases 1-6 (integrated, +/-0.1 LU), 9-14 (momentary/short-term dynamics and maxima, including the 20-step live sequences), 15-19 (true-peak tones, +0.2/-0.4 dB) and 20-23 (the fs/4 period inside an fs/6 tone decimated with 0-3 sample offsets); Tech 3342 LRA cases 1-4 (+/-1 LU) plus the repetition-invariance clause. Alongside: the 997 Hz / -3.01 LKFS anchor on L/C/R, the verbatim 48 kHz coefficient tables and the redesign match at 44.1/96/192 kHz, the Annex 2 Attachment 1 under-read bound, the Annex 3 Table 4 weight grid, LFE exclusion, gating behaviour and the validation/edge paths. The case tables live in tests/reference_data.py, shared with the conformance report. * feat: programme-loudness conformance checks (8 EBU cases) New 'Program loudness (ITU-R BS.1770 / EBU R 128)' domain in the conformance registry: the Annex 1 997 Hz anchor, Tech 3341 integrated cases 1/5/6, true-peak cases 15/19 and Tech 3342 LRA cases 1/3, all synthesized from the shared reference tables. 274 checks total. * docs: programme-loudness metering figure (EN/ES, light/dark) A one-minute synthetic programme (ambience, dialogue, music, fade-out) loudness-normalised to the R 128 target: momentary and short-term traces, the integrated loudness on -23.0 LUFS, the LRA band between its percentile edges and the summary box with the maxima and true peak. * docs: programme-loudness guide (EN/ES), indices and API reference The guide walks the chain from the K-weighting and the 997 Hz anchor through gating, the EBU Mode time scales, the loudness range, the true-peak meter and the Annex 3 multichannel weights, with the official EBU signals as inline examples and the validation scope spelled out. Wired into the documentation index, the sources-and-devices section landing (EN/ES), the sidebar, the curated API quick table (thirteen domains now), the generated API reference taxonomy, the bibliography and the changelog. * docs: add the programme-loudness guide to the README feature table * fix: address programme-loudness review findings - Reject sample rates below 16 kHz in the K-weighting redesign: at 8 kHz the bilinear warping reaches 0.55 dB in band and drifts the 997 Hz anchor to -2.89 LKFS, outside the +/-0.1 LU metering tolerance. The response claim is now scoped honestly (within 0.02 dB at 32 kHz and above) in the docstrings, the guides and the changelog. - Reject non-finite input in integrated_loudness/program_loudness: a single NaN would poison every gating block and silently read as digital silence (-inf). - Correct the EBU R 128 tolerance prose in the three guides: the 2023 edition permits +/-1.0 LU where the target is not practically achievable and +/-0.2 LU for measurement error in QC workflows (the +/-0.5 LU general tolerance was the 2014 rule). - Drop the false 'Annex 2 minimum of 4 below 96 kHz' clause from the true_peak_level docstring (the default is the smallest factor whose oversampled rate reaches 192 kHz) and regenerate the API page. - Reject a boolean oversample factor, make the second K-weighting snippet self-contained, drop an unused unpacked variable in the conformance checks and a redundant loudness-range guard. * test: run the authentic-programme EBU loudness cases against the official test set Tech 3341 cases 7-8 and Tech 3342 cases 5-6 use authentic programme material whose licence covers technical testing only, so the audio is never committed: the tests read the official EBU loudness test set from a gitignored local directory (EBU_LOUDNESS_TEST_SET overrides it) and skip when it is absent. One CI matrix cell fetches the set from the EBU with an Actions cache so the four cases also run on every pull request. All four pass within the official tolerances. * fix: harden the EBU test-set download and address the loudness review smells The EBU site rejects curl's default agent, so the CI download sends a browser agent, verifies a sentinel file and drops the directory on failure so a bad download is never cached (fresh cache key). The duplicated empty-signal message becomes a module constant and the exception tests build their inputs outside the raises blocks. * test: pin the gating and LRA stages to committed authentic-programme series The audio of the EBU authentic-programme cases cannot be committed, but the per-block loudness series measured from it are plain measurement data that cannot reconstruct the programmes. The 400 ms gate-input and 3 s short-term series of both files are committed (26 kB, rounded three orders of magnitude below the official tolerances) and the integrated-gating and loudness-range stages now verify the four case values on every platform, network or not. The full chain still runs against the original audio wherever the set is present. * ci: send a full browser agent when fetching the EBU loudness test set * fix: apply the programme-loudness review round The Tech 3341 true-peak conformance checks now judge with the asymmetric +0.2/-0.4 dB window the specification defines. Long programmes recover the inter-sample peak in bounded chunks (with a margin far beyond the FIR support, verified exactly equal to the whole-signal computation) so the memory use no longer scales with duration. Non-finite loudspeaker positions and channel weights are rejected, the momentary and short-term steps enforce the 10 Hz minimum update rate the API documents, the curated API table states the real 16 kHz floor, the plot's short-term label became a caller-overridable default, the download sentinel and the oracle skip check both files, the series test closes its npz handle and the Spanish landing names the exact BS.1770-5 revision.

2 months ago

README.md

EBU programme-loudness — derived validation data#

What this folder contains#

ebu_programme_block_loudness.npz holds block-loudness series measured from the authentic-programme cases of the EBU loudness test set (v04). They are used by the test suite to validate the gating and loudness-range stages of phonometry.broadcast.program_loudness against the EBU Tech 3341/3342 targets without any network access.

Array	Blocks	Source case	Content
`nlr_momentary`	400 ms / 100 ms hop	Tech 3341 case 7 (NLR narration)	Momentary loudness series, LUFS (the BS.1770-5 integrated-loudness gate input)
`nlr_short_term`	3 s / 100 ms hop	Tech 3342 case 5 (NLR)	Short-term loudness series, LUFS (the loudness-range input)
`wlr_momentary`	400 ms / 100 ms hop	Tech 3341 case 8 (WLR movie/drama)	Momentary loudness series, LUFS
`wlr_short_term`	3 s / 100 ms hop	Tech 3342 case 6 (WLR)	Short-term loudness series, LUFS

The series were measured with phonometry.broadcast.program_loudness from seq-3341-7_seq-3342-5-24bit.wav (NLR) and seq-3341-2011-8_seq-3342-6-24bit-v02.wav (WLR), then stored rounded to 0.0001 LU (three orders of magnitude below the official tolerances). They pin the integrated loudness to -23.0 LUFS and the loudness range to 5 LU (NLR) and 15 LU (WLR), the EBU-published targets for those cases.

Why only the derived series, not the audio#

Because the audio may not be redistributed, by two independent constraints.

The EBU's own Use of EBU audio test sequences (v1.0, July 2019) states: "You may not use any EBU Test Sequences for business, commercial or for-profit activities. You may not copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the EBU Test Sequences. All rights, titles and interests relating to any and all specific EBU Test Sequences belong to the EBU." The set's own readme.txt adds: "The material may only be used for technical testing purposes." That rules out committing any file of the set, not only the programme cases.

On top of that, cases 7/8 and 5/6 are licensed feature-film excerpts: the readme.txt credits The Misfortunates (dir. Felix van Groeningen, prod. Menuet, Belgium) and Bloody Mondays and Strawberry Pies (dir. Coco Schreiber, prod. Bonanza Films, the Netherlands), whose rights holders gave permission to the EBU, not to downstream users.

What is committed is therefore measurement data, not audio: a per-block loudness envelope at a 100 ms hop is our own computation and cannot reconstruct the programme signal.

Source and authorship#

Derived from the EBU loudness test set (v04), © EBU (European Broadcasting Union), https://tech.ebu.ch/publications/ebu_loudness_test_set (retrieved 2026-07-30; the page now offers v05).
Terms of use: https://tech.ebu.ch/files/live/sites/tech/files/shared/testmaterial/use%20of%20EBU%20AUDIO%20test%20sequences.pdf (Use of EBU audio test sequences, v1.0, July 2019), quoted above.
The target values reproduced by these series are defined in EBU Tech 3341 (loudness metering) and EBU Tech 3342 (loudness range), which in turn build on ITU-R BS.1770.
The .npz itself is an original transcription of numeric measurements, not an EBU file. Following the EBU's quotation clause: © EBU for the test sequences these measurements were taken from.

Purpose and scope of use#

The series are consumed by tests/broadcast/test_ebu_material_oracle.py to demonstrate that this library's independent gating and loudness-range implementation reproduces the EBU authentic-programme targets, everywhere including CI cells without network access. They are not part of the phonometry package and are not installed with it.

What they cannot cover is the chain upstream of the block envelope on authentic programme material: 24-bit WAV decoding, the BS.1770 K-weighting front end, the 400 ms / 100 ms segmentation and the channel weighting. The same module asserts those end to end from the original audio, at the same tolerances, wherever a local copy of the set exists (EBU_LOUDNESS_TEST_SET, or tests/data-local/ebu-loudness-test-set/); where the audio is there it also re-measures the envelopes and checks them against this file, so the derived oracle cannot drift away from the material it represents. Those stages are covered in CI by the synthesizable EBU cases in tests/broadcast/test_program_loudness.py, built from the specifications.

Removal policy#

If you represent the EBU and consider that publishing these derived series exceeds the intended use of the test set, please open an issue or contact the maintainer (see CITATION.cff) and they will be removed promptly, together with the tests/broadcast/test_ebu_material_oracle.py tests that read them.

Configure Feed