1 Introduction
Living reference. This book is versioned on Zenodo and revised on the web at voicemeasures.org. Each chapter cites primary sources with page anchors; findings you may want to check should be traceable to the exact page.
1.1 What this reference is
Acoustic Measures of Voice Quality is a living open reference on the acoustic analysis of the human voice. It is written for speech-language pathologists, voice researchers, and users of computational voice-analysis tools who need authoritative, citation-backed answers about what an acoustic measure means, how it is computed, what it says clinically, and where it fails.
The reference is deliberately narrow. It covers acoustic measures of voice quality — measures derived from the acoustic signal alone. Aerodynamic instrumentation, laryngeal imaging, and perceptual scales are in scope only where they inform how an acoustic measure is defined, trained, or validated. Perceptual scales such as GRBAS and CAPE-V appear because composite indices (AVQI, ABI) are trained against them; that treatment is in the Foundations section.
1.2 How the reference is organised
The reference is planned in five parts, ordered by what the measure captures rather than by clinical context. This release contains three Foundations chapters (signal fundamentals, signal typing, and pipeline dependence) and six measure chapters (jitter, shimmer, HNR, CPPS, AVQI and ABI). The other chapters are listed on the home page and are added as they are written. Chapters within each part follow a common template — definition, computation, normative values, clinical interpretation, caveats, and disagreements among sources when they arise.
- Part I — Foundations. Signal fundamentals, recording protocol, signal typing, F0 detection, voice tasks, perceptual anchors.
- Part II — Periodicity and stability. F0 measures, jitter variants, shimmer variants, voice breaks, and the shared perturbation-caveats discussion.
- Part III — Noise and harmonic structure. HNR and NHR, GNE, CPPS, spectral tilt.
- Part IV — Composite indices. AVQI, ABI, DSI, and the shared composite-caveats discussion.
- Part V — ML-derived measures. A landscape chapter plus two representative example measures (feature-based and end-to-end).
1.2.1 Which measure asks which question
Every measure in Parts II to IV is computed from the same thing — a recorded pressure waveform — and they differ in what they ask of it. The four written so far divide as follows.
| Measure | The question it asks of the waveform | Chapter |
|---|---|---|
| Jitter | Is each cycle the same length as the one before it? | Jitter |
| Shimmer | Is each cycle the same size as the one before it? | Shimmer |
| HNR | How much of the sound is periodic rather than noise? | HNR |
| CPP / CPPS | How strongly does a harmonic pattern stand out from the background? | CPPS |
| AVQI | Combining several of the above, how dysphonic does the voice sound? | AVQI |
| ABI | Combining several of the above, how breathy does the voice sound? | ABI |
Jitter and shimmer are cycle-by-cycle measures: they need the cycles to be found first, which is why they are confined to nearly periodic signals. HNR and CPP ask about the signal as a whole, and CPP in particular needs no cycle boundaries at all. That difference governs which measure survives in a disordered voice, and it is taken up in each chapter.
Back matter includes appendices on adjunct methods (EGG and aerodynamic measures at the level they inform acoustic interpretation), reproducibility notes for Praat and PhonaLab, and a glossary.
1.3 How to use it
Every chapter is self-contained and cites the primary sources. Three reading paths:
- Clinician looking up a specific measure. Go straight to the measure chapter — the Definition, Normative Values, and Clinical Interpretation sections are what you need. The Caveats section is where the honest cautions live; read it before reporting a value.
- Researcher tracing methodology. The Computation section names the algorithm variants and software; the reference list gives the primary papers.
- PhonaLab user. Each measure chapter links to PhonaLab, which computes the measure. The chapters explain what the number means and which reference values apply to it. See Competing interests below.
Every factual claim carries a source. Where sources disagree, both are shown and the disagreement is called out — never silently averaged.
1.4 Versioning, DOIs, and citation
Each release is deposited on Zenodo with a versioned DOI. Cite this reference as, for example:
Lucero, J. C. (2026). Acoustic Measures of Voice Quality. Zenodo. https://doi.org/10.5281/zenodo.22739049
Where a specific chapter is cited, include the chapter title and section. To cite a specific release, use the versioned DOI on that release’s Zenodo record.
1.5 Competing interests
The author of this reference, Jorge C. Lucero, develops PhonaLab, a commercial web application for acoustic voice analysis. The reference is open and independent of it. Every reference value here comes from the published sources cited, with its pipeline and compatibility class. No value is chosen, kept or omitted because of how it compares with PhonaLab’s output, and PhonaLab is not treated as a source. Each measure chapter ends with a link to PhonaLab, labelled as the author’s software.
1.6 Licence
The text is licensed under CC BY 4.0; any computational code shown in the reference is licensed under the MIT License. Attribution: Jorge C. Lucero, University of Brasília.
1.7 Contributing and errata
Corrections, additions, and disagreements are welcome. The reference is maintained on GitHub at github.com/LuceroJC/voice-measures — open an issue with a specific chapter and page, and where possible cite a source. Contributed sections are credited in the changelog.
Corrections to values or statements that appeared in an earlier release are recorded in Release notes and errata, with what the earlier release said and what replaced it. Corrections are not made silently.
The reference is written to be updated. Findings that survive on the site for a full release cycle without correction can be treated as verified; findings marked as draft are still under review.