Release notes and errata

This reference is written to be revised, and each release is deposited on Zenodo with its own versioned DOI. When a value or a statement that appeared in an earlier release is corrected, the change is recorded here, so that a reader working from an earlier version can see exactly what moved and why. Corrections are not made silently.

Entries are newest first. “v0.1.0 said” refers to the text as deposited in that release’s Zenodo record, not to the live site, which is revised continuously between releases.

v0.1.1 — 2026-10-06

A correction and sourcing release. No chapter was added or removed. Eight statements published in v0.1.0 are corrected, five of them because the claim has since been checked against its primary source rather than a secondary citation; four new language validations are added.

Corrections to v0.1.0

Signal Typing — @tbl-type-measures, Type 2 row. v0.1.0 followed Ludlow, Kent and Gray on what may be measured from a Type 2 signal. The table now follows Titze’s summary statement [p. 19, pp. 26–27], which differs. This is a substantive change to what the table says about Type 2, not a rewording. The disagreement between Titze and Sprecher et al. on whether Type 2 signals support perturbation analysis is now stated explicitly in Points of disagreement, together with the authors’ own acknowledgement of it.

AVQI — Dutch internal validation. v0.1.0 carried the sensitivity and specificity as reported in the secondary literature, which transposes them. The correct figures are sensitivity 1.00 and specificity 0.936 at a threshold of 2.43, AROC 0.98. The figures are now also correctly attributed, to Barsties v. Latoszek and Maryn (2015), VD-2 condition, p. 653, and identified as development-sample figures. They are therefore recorded as internal-validation context and not as a second Dutch cutoff; the Dutch row keeps its external validation on 1058 samples.

AVQI — Japanese row. v0.1.0 carried the threshold of 2.06 from a second-hand citation, without the pipeline or the derivation criterion. The value is confirmed against the primary source (Hosokawa et al. 2019, Journal of Voice 33(1):125.e1–125.e12): 2.06, sensitivity 0.721, specificity 0.938, AUC 0.915, 30 syllables, 455 recordings. The derivation criterion is now stated, and it matters: 2.06 is the 95th percentile of the study’s controls, not a Youden-index value. The same paper derives 1.41 by the Youden index and reports 64.6% / 97.3% for the Dutch 2.43 applied to its own sample. The authors recommend 2.06 and the table follows them, so the Japanese row is not criterion-comparable with the Youden-derived rows.

AVQI — threshold caveat. The caveat that thresholds differ “by language” now reads by language, by version, and by derivation criterion. Part of the published spread in AVQI thresholds is a spread of derivation methods, not of populations.

ABI — Japanese row. The row is promoted from a second-hand citation to its primary source (Hosokawa et al. 2019, JSLHR 62(8):2617–2631): 3.44, sensitivity 76.3%, specificity 94.1%, AUC 0.939, Praat 6.0.20, 30 syllables. Three easily confused Hosokawa papers are now distinguished in the chapter — the AVQI v02.02 validation, the AVQI v03.01 validation, and this ABI validation.

ABI — Brazilian Portuguese provenance. v0.1.0 cited this threshold through a meta-analysis, as “Int Arch Otorhinolaryngol 2019;23:106”. That citation does not resolve. The source is abstract 8972 in the annals of the 18th Congress of the Otorhinolaryngology Foundation, p. S106, and is now cited directly. The meta-analysis’s Table 1 superscripts are off by one on four of six rows; this is recorded so that readers checking the meta-analysis against this reference are not misled by it.

ABI — evidence-base caveat. The Fleiss κ of 0.353 is now cited directly to the abstract rather than to the meta-analysis, and identified as the breathiness agreement figure specifically, not an overall one.

CPPS — history table. The Hillenbrand-derived cutoffs were published without their program recorded. They are SpeechTool-scale values. Praat and SpeechTool CPPS differ by about 5 dB on the same 289 recordings (6.61 against 11.66), with the relation y = 1.4387x + 2.1543, r² = 0.924, so a cutoff from one program cannot be applied to output from the other. This explains why the Hillenbrand cutoffs are not low Praat values. The ADSV rows are unaffected by this correction; they remain unclassified for a separate reason, namely that no ADSV version is stated by their sources.

Added in v0.1.1

  • ABI, Brazilian Portuguese: 2.94 — counting 1 to 20 with the voiced parts equalized (37 syllables), 150 speakers; sensitivity 75.3%, specificity 93.4%, ROC accuracy 85.2%.
  • ABI, Malayalam and Kannada: 3.66 — the breathiness-referenced value. The 3.41 reported in the same study against overall grade is recorded only to prevent its misuse. The study’s two printed LR− values are impossible as printed; recomputed they are 0.40 and 0.51.
  • ABI, Persian: 2.97. The published ABI thresholds now span 2.35 to 3.69.
  • AVQI, Persian: 1.63.
  • AVQI, Japanese v02.02: 3.15 — on a different scale from the v03.01 value, and distinguished from it.
  • AVQI: a new subsection on threshold derivation criteria, and a caveat on the equal proportion required of the two speech tasks.
  • Signal Typing: Type 4 re-sourced to Sprecher et al. (2010), including the point that adding Type 4 subdivides Titze’s Type 3 rather than contradicting it.

Declared absences

Two rows carry no Praat version, because none is stated by the source:

  • the Brazilian Portuguese ABI row (2.94) — the congress abstract does not state one; the companion paper on the same sample and task states 6.0.40, which is recorded as context and not as this row’s pipeline;
  • the Japanese AVQI v03.01 row (2.06).

These are recorded as not stated rather than inferred. A pipeline that the source does not state is not supplied from elsewhere.

Held

Normative AVQI data for Tamil is held out of the chapters until a paginated version of record exists. Its reported mean of 3.4 in normophonic speakers sits above every published v03.01 dysphonia threshold, which points to its speech material — seven whole CAPE-V sentences rather than a standardized syllable count — but that objection is an inference rather than a statement in the source, and the source carries no pagination to anchor it to.

v0.1.0 — 2026-09-29

First public release. Three measure chapters — Cepstral Peak Prominence (CPP and CPPS), the Acoustic Voice Quality Index (AVQI), and the Acoustic Breathiness Index (ABI) — with three Foundations chapters on signal fundamentals, signal typing, and pipeline dependence.

Citing a specific version

The concept DOI 10.5281/zenodo.22739049 always resolves to the latest release. To cite the version you actually read, use the versioned DOI shown on that release’s Zenodo record, and give the chapter title and section:

Lucero, J. C. (2026). Acoustic Measures of Voice Quality, v0.1.1, “Acoustic Breathiness Index”, §7.3. Zenodo. https://doi.org/10.5281/zenodo.23190550