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 / stipa /

at main 3 files

José M. Requena Plens Say the message once, in the exception itself (#607) 7w ago

8fa0e316

README.md

Split hearing into speech and hearing (#473) * Split hearing into speech and hearing The package held two subjects that share a word and nothing else. Speech intelligibility scores a transmission channel and is read by whoever designs a room or a public address system; hearing conservation is about the ear and is read by an occupational hygienist. The generated reference and the sidebar had separated them long ago, which is why the reference had a speech section drawing from a package called something else; the code is what had not caught up. STI, SII and STOI become phonometry.speech. The audiometric thresholds, the ISO 1999 hearing loss and the ISO 9612 exposure stay in phonometry.hearing, which is now what its name says. No module is renamed and no reference page moves, so this is the cheapest of the taxonomy splits and the one that removes an exception rather than adding a URL. Compatibility follows the pattern the metrology split established: the three pre-split module paths still import and delegate with a notice, reading a moved name from the namespace it left still works, dir() still lists them, and all of it goes in 5.0. The frozen snapshot of pre-split paths grows by three. The renderers split the same way, so _plot/hearing.py becomes two modules with their own translation tables. The speech half gets the Spanish test it never had, including the four vocal efforts of the ANSI S3.5 spectra, which are table keys reached through the result rather than written in the renderer. The snippet gate earned its place here: it caught the SII guide importing the new speech package into a block that already bound `speech` as the variable holding the speech spectrum. The variable is renamed. * Point the last callers at the packages, and widen the snippet gate From an adversarial review of the split. The fiche generator still called the library through the namespace the names left: `ph.hearing.sti_from_impulse_response` and its SII neighbour, the two factories behind the STI and SII example fiches. They ran on a deprecation notice nothing turns red, and they would have stopped running in 5.0. The sibling scripts were migrated with the split; these two were missed. The snippet gate looked at the guides and at nothing else, which is where the one broken example was hiding: `llms.txt`, the first thing an assistant reads, opened with `from phonometry import filters, signal` and the package has been `signals` since the rename. The gate now covers the README, the PyPI long description and `llms.txt` the same way it covers a guide, and it reads the generated dumps too. Those dumps glue unrelated pages into one file, so they are read block by block: carrying an import across a page boundary invents collisions no reader can hit, which is what the first run reported. A block that passes `...` to a call is a sketch of a call, not a computation, so it is read but not run; the README leads with one. Three links pointed at `src/phonometry/hearing/sii.py`, which is now under `speech`, from the errata page in both languages and in the GitHub mirror. Two test-data READMEs cited test files that moved, one of them in this split and one back in the 3.2 modularization. The Spanish test for the speech renderers claimed to hold the translation table to its contents and did not: deleting three of the four band-importance labels left it green, because it exercised one procedure and never read the legend. It now checks all four, and a mutation of the table fails it. The CHANGELOG said three sections carry a key that is not a subpackage name and then listed four things, one of which is itself three. It is five, and they are named. * Answer the review comments Every Sphinx cross-reference in the package that named a pre-4.0 module path now names the module it points at. There were about ninety of them, nearly all in the renderers, and two were the ones the reviewer flagged: the speech renderers pointed at `phonometry.sti` and `phonometry.sii`, which are 3.1 aliases that warn. They are resolved from the alias table itself rather than by hand, so the class is gone rather than the two instances. The table's own file is left alone: naming the deprecated paths is its subject. The release note listed STI, SII and STOI and left out STIPA and ESTOI, which move with them and are public API of their own. A call sketch written with a keyword placeholder, `f(level=...)`, is the same thing as `f(...)` and the gate now reads it as one instead of running it. The repeated axis labels in the new speech renderers become module constants, which is what SonarCloud asked for and what the metrology split already did. The two branches the coverage report named are now tested rather than excused: the modulation-index axis of an STI result that does not carry the seven octave bands, which a caller can build because the result class is public, and the refusal to plot an exposure result that carries no per-task contributions.

2 months ago

make_extract.py

Say the message once, in the exception itself (#607)

1 month ago

stipa_certified_extract.zip

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

README.md

IEC 60268-16 STIPA certified verification bench — extract#

What this folder contains#

stipa_certified_extract.zip holds 27 of the 49 signals of the certified STIPA verification bench published on stipa.info, the test signals IEC 60268-16:2020 (rev 5) Annexes A and C prescribe for verifying a STIPA implementation. They are consumed by tests/speech/test_stipa_certified.py.

Verification suite	Committed	Of	Why these
`Annex C.3.2` direct-method modulation depth	3	11	`m = 0` (null), `m = 0,9` (worst measured STI deviation of the staircase, 0,0031) and `m = 1` (top of the dynamic range, worst measured MTF deviation, 0,0065)
`Annex C.3.3` indirect-method modulation depth	7	7	the complete suite: two pure decaying carriers per file encode to a few tens of kilobytes
`Annex C.4.2` filter-bank slope	14	14	the complete suite, for the same reason; this is the criterion that forces an effective skirt attenuation above 41 dB
`Annex A.2.2` weighting/redundancy factors	2	6	the lowest (`125+250`) and the highest-STI (`1000+2000`) band pair
`Annex A.3.1.2` filter-bank phase distortion	1	11	`TI = 0,9`, the worst measured phase bias (-0,0029) inside the normative `TI = 0,1 .. 0,9` range

The 49 WAV files are 133 MB of 48 kHz 16-bit PCM which does not compress as audio, so committing the bench as WAV is out of the question. Each signal here is stored losslessly: the n-th order difference of its int16 samples as little-endian int32, LZMA-compressed inside the zip, with manifest.json recording the order, the sample rate, the length and the SHA-256 of the original samples. Decoding is n cumulative sums, and the test suite checks the digest on every load, so what the tests analyse is bit-exact with what stipa.info publishes. Only the 44-byte RIFF header of each WAV is not reproduced; the audio samples are.

make_extract.py rebuilds the archive from a local copy of the full bench:

python tests/data/stipa/make_extract.py tests/data-local/stipa-verification

Source and authorship#

The verification test signals were developed and are published by Embedded Acoustics BV (Jan Verhave), Delft, the Netherlands, on the download page of https://www.stipa.info, together with the signal description IEC-60268-16 revision 5 test signal description, v1.0, 3 June 2020.
Authoritative record of the terms (cite this one): Internet Archive capture of stipa.info/index.php/download-test-signals taken 2025-09-10, https://web.archive.org/web/20250910080127/https://www.stipa.info/index.php/download-test-signals. The live page at https://www.stipa.info/index.php/download-test-signals carries the same text but is currently defaced (checked 2026-07-30: it serves an injected script that rewrites the document to "Hacked By Simsimi"). A licence claim should not rest on a page an attacker has had write access to, so the archived copy is the citation of record here. The Archive holds captures from 2022-04-23 to 2025-09-10, all predating the compromise; the terms below are identical in the first and the last of them.
Terms, quoted from that capture: "All of the above verification test signals were developed by Embedded Acoustics and may be used for commercial and non-commercial use free of charge." The same page is explicit that the signals are not STIPA measurement signals and that the STIPA measurement signals sold by instrument manufacturers are separately copyrighted; none of those is used here.
What that grant does and does not say. It grants use free of charge, for commercial and non-commercial purposes alike. It does not use the word "redistribute" about the verification signals — but the page shows that its author treats redistribution as a distinct permission and attaches an explicit condition wherever one is intended: about the manufacturers' measurement signals it says "Bedrock allows unrestricted non-commercial use of their signals, but a paid license is required if you intend to redistribute or promote the signal for use with your own STIPA implementation." No such condition is attached to the verification signals.
This redistribution rests on our own reading of those terms, not on a permission granted to this repository: an unconditional free-of-charge grant, from an author who conditions redistribution where he means to, over material published precisely so that implementations can be verified against it, is read here as covering a subset carried inside a verification suite. If Embedded Acoustics reads it otherwise, the removal policy below applies and the suites fall back to the local-only bench.
The encoded blobs in the zip are a lossless re-encoding of that material, not original Embedded Acoustics files; manifest.json is ours.

Purpose and scope of use#

Embedded Acoustics publishes the bench so that STIPA implementations can be verified against the criteria of IEC 60268-16 Annexes A and C. The extract in this folder serves exactly that purpose: it lets tests/speech/test_stipa_certified.py demonstrate that this library's independent implementation meets those criteria on every runner, rather than only on a machine that has the full download. The files are not part of the phonometry package, are not installed with it, and are not covered by this repository's MIT licence — the free-of-charge terms above apply to them.

Deliberately excluded: the bundled reference .m Matlab sources (the library is written from the standard, not from them), the original zip archives, and the signal-description PDF. The 22 bench signals that are not committed are the intermediate points of the C.3.2 staircase, four of the six A.2.2 band pairs and ten of the eleven A.3.1.2 phase points; they are the most costly to encode (nearly full-entropy multi-tone signals) and each is a further point on a curve whose ends and worst case are already committed. They run automatically when a full local copy is present.

Removal policy#

If you represent Embedded Acoustics BV and consider that committing this extract exceeds the intended use of the verification bench, please open an issue or contact the maintainer (see CITATION.cff) and it will be removed promptly, together with the CI use of the tests that read it.

Configure Feed