Release notes

0.9.5 (18/05/2026)

Sound and annotation views

  • Mouse-wheel navigation now follows the convention used by recent versions of Praat and wavesurfer.js: a plain scroll over the waveform, spectrogram, pitch track, intensity track, or annotation layers pans the visible window left (scroll up) or right (scroll down). Zooming has been remapped to Ctrl+Scroll (Cmd+Scroll on macOS), which still zooms around the time under the cursor. The wave bar’s pan-on-scroll behaviour is unchanged.

  • Annotation file operations in the file manager context menu: Duplicate, Extract layers…, Extract slice…, Merge annotations…, and Concatenate annotations…. Each produces a new file on disk and adds it to the project, inheriting properties, description, and (where preserved) sound binding from the source(s). Merge requires matching durations within 1 ms; concatenate accepts per-source explicit durations for unbound inputs. The new annotation keeps the source’s format by default (native .phon-annot or Praat TextGrid), with a selector when sources disagree.

  • Sound file operations: Extract slice… (single sound or as the sound side of an annotation slice) and Concatenate sounds…. Concatenation is strict-mode (matching sample rate and channel count); resample-on-concatenate is planned. Output format (WAV, AIFF, FLAC, OGG) is selectable.

  • Extract slice… on a bound annotation offers three modes — annotation only, sound only, or both; the both mode binds the new annotation to the new sound automatically.

Acoustic queries

  • Voice quality query (Analysis > Measure voice quality...): batch-mode equivalent of the interactive Voice report (F9 in the sound view). Runs a text search and then computes jitter, shimmer, HNR, and pulse-summary measures on each matched interval. The dialog exposes the F0 range and a 14-checkbox grid for selecting which measures appear as concordance columns, with Select essentials (pulses, mean F0, jitter local, shimmer local, HNR) and Select all presets. Measurements are stored in display units (ms, %, µs, dB, Hz, integer pulses) so CSV exports are immediately readable. Voice quality is whole-interval only — there is no midpoint or n-point option — and instant targets are skipped with a single summary warning. Headers mirror Praat’s naming.

Scripting

  • List comprehensions for building lists in expression form: [expr foreach var in coll], with optional if cond filter and if cond else other_expr conditional clauses. The two-variable form foreach k, v in coll supports tables and indexed iteration over lists. Lowered directly to a foreach-shaped loop over a hidden accumulator (no closure overhead).

  • New globals matching the GUI operations: duplicate_annotation, extract_layers, merge_annotations, extract_annotation_slice (two overloads), concatenate_annotations (two overloads), extract_sound_slice, concatenate_sounds. All return a fresh handle to a file written on disk; the caller decides whether to import it into the project.

  • improved scripting editor: better error reporting, go to definition, auto-complete and call tips for locally defined symbols

0.9.4 (14/05/2026)

General

  • New Fitting options popup (gear icon next to the Fit button in the analysis view) gathers settings that apply to the next fit: default estimation type (frequentist / Bayesian), maximum optimizer iterations, and — for Gaussian linear mixed models — the choice between ML and REML.

Statistical analysis

  • Restricted maximum likelihood is now available as an opt-in alternative to maximum likelihood for Gaussian linear mixed models.

  • improvements to the statistical engine

Exploratory data analysis

  • Group and Facet variables are now available across all EDA plot types (Histogram, Bar chart, Boxplot, Scatter, Formant chart). Group splits the data within each panel into colored subseries — overlaid translucent fills for histograms, dodged sub-bars for bar charts, side-by-side boxes for boxplots, and color (plus marker shape via Style) for scatter plots. Facet builds small multiples, one panel per level of a categorical variable; panels are laid out near-square with a four-column cap by default, overridable via the Customize dialog. Group and Facet combine freely.

  • Pool by (scatter and formant chart, with Group set): average X and Y values within each (group, pool) cell before plotting — for example, pool by speaker to obtain one point per speaker per vowel. This restores the pre-grouping behaviour of formant charts as an explicit, named option.

  • Style (scatter and formant chart, with Group set): encode a second categorical variable as marker shape (and confidence-ellipse line style), freeing colour to encode Group.

  • New Customize… dialog for EDA plots gives access to the plot title, X / Y / legend titles, X and Y range overrides (with comma- or dot-decimal separators), and the per-row facet column count. Customizations persist across variable changes but reset automatically on plot-type change. A Reset button clears all customizations at once.

  • The plot type can now be forced manually (Histogram / Bar chart / Boxplot / Scatter / Formant chart) instead of always being inferred from the selected variables, with an inline hint when the chosen variables don’t fit the chosen plot type.

Internal

  • New R-INLA-based validation suite (test/statistics/bayesian/inla/): four chapters of brinla reference fits (BLR / GLM / GLMM / GAM) with reference posteriors stored as JSON and matching .phon test scripts.

  • New REML validation suite against lme4 / glmmTMB reference fits.

0.9.3 (10/05/2026)

General

  • All plots in analysis views are now detachable into their own window.

  • Save dialogs for analyses propose a default filename derived from the source data table.

  • The user manual now opens in the system web browser instead of the embedded Qt help viewer; equation rendering in the documentation has been improved.

Sound and annotation views

  • New audio recording capability (WAV PCM_16, unbounded duration), accessible from the main window.

Statistical analysis

  • Nested random effects are now supported, in either of two equivalent forms: lme4-style slash sugar (1 | school/classroom), which expands at parse time to (1 | school) + (1 | classroom:school); and explicit bare-colon synthetic groups (1 | g1:g2), which build a grouping factor over the observed pairs of values. Slash sugar follows the lme4 convention with the inner factor on the left of the synthetic name; the bare-colon form preserves the order written. The right-click menu in the analysis view now offers an Add as nested grouping factor in… submenu listing existing grouping factors in the formula.

  • Predicted effects for fixed-effects, mixed-effects, and Bayesian models. Mixed-effects predictions are available in both conditional and population-averaged forms.

  • Posterior predictive checks for Bayesian models, with a residual-diagnostics tab using frequentist DHARMa-style p-values for cross-family consistency.

  • Refit model action re-runs the fit with the current settings without reopening the analysis.

  • Improved accuracy of frequentist Student-t (robust) regression: multi-start optimisation, exact-Hessian Laplace correction with Fisher-information fallback when the Hessian is non-positive-definite.

  • Improved accuracy of Bayesian negative binomial models and of Bayesian robust regression in both fixed-only and mixed forms.

  • Improved GAM accuracy. GAMs are now explicitly marked as experimental in the documentation pending further validation.

  • Validation suites against glmmTMB (frequentist) and brms (Bayesian) reference fits for the negative binomial, Student-t, and Bayesian Gaussian families.

  • EDA scatter plots: the Regression line option now stays available when a grouping variable is selected. With grouping, one OLS line is drawn per group, in the group’s colour and clipped to that group’s own x-range; it can be combined freely with the existing means and confidence ellipses.

  • Clicking an observation in an EDA or residual-diagnostics plot opens the corresponding row in the source data table.

Scripting

  • New get_script_path() builtin returning the absolute path of the currently-running script.

  • Numeric literals can now be written in scientific notation (e.g. 1.5e-3).

  • Syntactic sugar for named arguments: positional arguments are passed first, followed by key = value pairs that are collected into a single Table argument.

  • Fixed compound assignment on Module fields (e.g. module.x += 1).

  • VM instruction format widened to 16-bit operands, raising the per-function limits on constants, locals, and arguments.

Internal

  • Updated bundled r8brain resampler to the latest upstream version.

0.9.2 (04/05/2026)

General

  • Support for drag’n drop in a running instance

  • Save dialogs propose the document’s label as the default filename, falling back to untitled when no label has been set.

Sound and annotation views

  • The mouse cursor is now tracked across linked sound plots while dragging an anchor on an annotation layer.

Concordance views

  • The tab title now reliably shows the modified-state asterisk for all concordance edits — editing match text, editing the underlying annotation event, toggling the Wide/Long layout, and cell-level changes via undo/redo.

  • Fixed a bug whereby filter rules could appear duplicated (sometimes many times over) after reopening a project, and where clearing a filter via the toolbar did not persist across reopens. Filter rules are now stored exclusively in the project file; existing concordance files with stale or duplicated rules are silently cleaned up the next time they are saved.

  • Fixed row removal in concordances containing auxiliary columns.

Statistical analysis

  • Improved accuracy and speed of binomial (logistic) regression, with further improvements to Bayesian logistic models that include random slopes.

  • Improved accuracy of robust (Student-t) regression in both frequentist and Bayesian fits.

  • Improved accuracy of Bayesian negative binomial models.

  • Bayesian models now report a posterior summary for random-effect correlations.

  • Random-slope terms can now contain interactions (e.g. (1 + x:y | g)).

  • Variables with parentheses in their names are now properly quoted in formulas; main effects and interactions are kept in shorthand form (e.g. x*y rather than x + y + x:y) in formula summaries.

  • Column-width formatting in model summary tables now adapts to the longest coefficient or hyperparameter name, so long names no longer push value columns out of alignment.

Speech

  • Audio transcription no longer offers a translation option; transcription is performed in the source language only.

Scripting

  • New try ... catch ... end block to catch errors to catch errors thrown by Phonometrica or by a user script with throw.

  • New Python-style triple-quoted string literals ("""...""" and '''...''') for writing multi-line strings without manual concatenation. Line breaks and isolated occurrences of the delimiter are part of the content; the string is closed by three delimiter characters in a row. Escape sequences are processed as in single-line strings.

  • Single-quoted and double-quoted strings now report a syntax error when a literal line break is encountered before the closing delimiter. The error message points to the triple-quoted form. This matches the behaviour of most mainstream scripting languages and is a one-time, intentional break in compatibility with previous versions.

  • New get_column(datatable, name) to retrieve a column by name from a dataset or concordance, complementing the existing index-based form, as well as append(datatable, column, name) to add a column to a data table.

  • Additional model-checking accessors exposed for analyses (fixed-effect inference, coefficient and hyperparameter names, hyperparameter posteriors, random-effects summaries).

  • Documents (annotations, concordances, datasets, notes, scripts, sounds, spectra) are now explicitly non-clonable.

0.9.1 (24/04/2026)

This release features usability improvements and bug fixes.

General

  • Settings dialogs for formants, pitch, intensity, and spectrograms are now non-modal: you can keep them open while interacting with the sound view and see parameter changes take effect immediately.

  • macOS: fixed communication with Praat when launching files from the file manager, annotation views, and concordance views.

  • macOS: toolbar separators in sound views are now drawn reliably, matching the appearance on other platforms.

Sound and annotation views

  • Rewritten pitch-tracking algorithm (Praat backend) for improved accuracy and robustness across voice types, with additional parameters exposed in the pitch settings dialog.

  • New keyboard shortcuts for acoustic measurements under the cursor or within a selection: F5 / Shift+F5 for FFT and FFT+LPC spectra; F6 / Shift+F6 for formants and mean formants; F7 / Shift+F7 for pitch and mean pitch; F8 / Shift+F8 for intensity and mean intensity.

Concordance views

  • Base cells (match text and auxiliary columns) are now editable inline: double-click a cell, or use the column’s context menu, to edit it in place. Edits participate in the undo/redo stack.

  • New keyboard shortcuts in the concordance table: Space plays the selected match, Esc stops playback, and Ctrl+Return opens the match in its annotation.

  • Protocol queries: split-field output (one column per protocol field in the concordance) is once again controlled by the coding protocol, matching the documented behaviour.

  • Fixed a bug whereby imported concordance files were filed under the Scripts folder instead of Concordances in the project tree.

  • Fixed serialization of NaN values in concordance and dataset files, which could prevent re-opening documents containing missing measurements.

Bookmarks

  • Creating a bookmark from a concordance now opens a dedicated editor where you can set the bookmark’s title and add free-form notes. The title is used as the label in the file manager.

  • Bookmarks can now be created from annotation views: select an event on a focused layer and click the new bookmark button in the toolbar, or press Ctrl+B. The selected event’s text seeds the default title.

  • Hovering over a bookmark in the file manager now displays a tooltip showing the file, the match (with KWIC context for concordance bookmarks, or the layer and time span for annotation-view bookmarks), and any notes you entered.

Scripting

  • Fixed String::to_float() for a class of numeric strings that previously produced incorrect conversions or failed to parse.

0.9.0 (17/04/2026)

This is a major release. Phonometrica has gained substantial new functionality for acoustic analysis, data exploration, and statistical modeling.

General

  • New user interface based on Qt 6.

  • Start view with quick-access buttons (Open Project, Add Files, New Annotation, Analyze Data, Documentation) and a recent projects list.

  • Preferences dialog with three tabs (General, Measurement, Appearance): configure Praat path, default estimation method, query context settings, display precision, and monospaced font.

  • IPA panel for inserting phonetic symbols into annotations and scripts.

  • Output panel for script execution output, separate from the interactive console.

  • Batch property editing: select multiple files in the file manager and edit their properties at once through the information panel, with a union property table showing coverage counts.

  • Context-sensitive help: each view has a Help button that links to the relevant documentation page.

  • Research notes: create and edit rich-text notes (bold, italic, headings, lists) within your project, stored as HTML (.phon-note).

  • Praat integration: configure the Praat path in preferences; “Open in Praat” from the file manager, annotation views, and concordance views; auto-detection of Praat on macOS and Windows.

  • Bidirectional TextGrid ↔ native annotation conversion with property, description, and sound binding transfer.

Sound visualization

  • Spectral slice (power spectrum): view the FFT and/or LPC spectral envelope at the cursor position; export to PNG, PDF, or SVG.

  • Spectral moments: compute centre of gravity, spread, skewness, and kurtosis at the cursor position or within a selection.

  • Plot export improvements: waveforms, spectrograms, pitch tracks, and other plots can be exported with Retina-quality rendering.

Queries

  • New query relation types: left alignment, right alignment, strict dominance, precedes, and follows, in addition to the existing alignment, dominance, and precedence.

  • Formant queries: extract formant measurements (F1–F5 and bandwidths) at the midpoint or as n-point averages; automatic formant selection using Weenink’s method; on-the-fly ERB and Bark conversion.

  • Pitch queries: extract pitch (F0) measurements using one of five algorithms (REAPER, Harvest, RAPT, SWIPE, Praat); semitone and ERB conversion.

  • Intensity queries: extract intensity (dB) measurements at the midpoint or as n-point averages.

  • Spectral moments queries: extract centre of gravity, spread, skewness, and kurtosis from the power spectrum; configurable window duration, window type, frequency range, and pre-emphasis.

Concordance views

  • New concordance view with toolbar for playback, annotation viewing, bookmarking, editing, and CSV export.

  • Set operations on concordances: union, intersection, and complement.

  • Horizontal merge: add columns from another concordance or dataset.

  • Column-level operations: rename, sort, recode (categorical), and transform (numeric).

  • Filter bar: define filter rules on any column (text or numeric) with operators such as equals, contains, matches regex, is one of, comparison operators.

  • Subset creation from filtered rows.

  • Metric columns for outlier detection: z-score, modified z-score, Mahalanobis distance.

  • Wide/long format toggle for n-point acoustic data.

  • Scales menu for toggling ERB, Bark, and semitone columns.

  • Vowel normalization: Lobanov, Nearey 1, Nearey 2, and Watt & Fabricius methods, accessible from concordance and dataset views.

  • Target highlighting (bold red).

  • Split view: option to open annotations beside the concordance.

Dataset views

  • New dataset view for tabular data (CSV import).

  • Same filtering, set operation, merge, recode, transform, and metric column features as concordance views.

  • Column operations: rename, duplicate, move, delete.

Statistical analysis

  • Analysis view: visual workspace for fitting, comparing, and diagnosing statistical models.

  • Visual formula builder with right-click context menus on columns.

  • Fixed-effects GLMs: Gaussian, binomial, Poisson, and negative binomial (NB2) regression.

  • Beta regression (logit link) for proportion response variables.

  • Student t regression (identity link) for robust estimation with heavy-tailed residuals; jointly estimates scale σ and degrees-of-freedom ν parameters.

  • Mixed-effects models (LMM/GLMM): support for random intercepts and correlated random slopes; crossed random effects.

  • Generalized additive models (GAM): support for penalized cubic regression splines, per-smooth GCV, by-variable smooths, EDF and F-statistics.

  • Random intercepts and random slopes via s(group, bs=re) and s(group, by=x, bs=re) in GAMs.

  • Model comparison: AIC/BIC table, pairwise likelihood ratio tests with automatic nestedness checking.

  • Approximate Bayesian inference (INLA-style): data-dependent weakly informative priors, posterior summaries (mean, credible intervals, probability of direction), customizable priors.

  • Bayesian model comparison: WAIC, PSIS-LOO-IC with Pareto k diagnostics, log Bayes factors.

  • Estimated marginal means (EMMs) and pairwise contrasts with Holm/Bonferroni adjustment; emtrends mode for testing slopes across groups. Bayesian EMMs report credible intervals and probability of direction.

  • Reference level setting for categorical variables.

  • DHARMa-style simulation-based scaled residuals for all model families.

  • Pseudo R² (Nakagawa & Schielzeth) for mixed-effects models.

  • Diagnostic plots: residuals vs. fitted, Normal Q-Q, scaled residual plots with Kolmogorov–Smirnov, dispersion, and outlier tests.

  • Exploratory data analysis (EDA): scatter plots, histograms with kernel density, grouped strip charts; confidence ellipses; formant chart mode (reversed axes); pooled means; regression lines; detachable plot window.

  • Summary export to plain text, clipboard, and LaTeX.

  • Analysis documents saved in XML format with full-precision coefficients.

Column transformations

  • Transform dialog with real-time preview: apply mathematical formulas to create new numeric columns.

  • Built-in phonetic scale functions: bark(), erb(), mel(), st().

  • Standard math functions: log(), log10(), log2(), sqrt(), abs(), exp(), pow(), round(), floor(), ceil().

Plugins

  • Plugin management from the GUI: install and uninstall plugins via the Plugins menu.

  • Plugin actions appear as submenus in the Plugins menu.

0.8.0 (2020)

This is an internal development version that was not released publicly.

  • new scripting engine

  • faster hash table based on Robin Hood hashing

  • save annotation in annotation view with Ctrl+S

0.7.6 (18/05/2020)

  • properties are now displayed in a table in the information panel

  • better error reporting

  • bug fixes

0.7.5 (09/11/2019)

  • n-point average in formant analysis

  • ability to remove rows in query views

  • fix navigation with arrows in annotation views when some layers are hidden

  • return empty label instead of crashing in formant analysis when intervals are misaligned

  • get_selected_annotations() and get_selected_sounds() functions

0.7.4 (14/11/2019)

  • robust sandwich variance estimator for Poisson regression

  • fix project finalization when views are modified

  • covrc() was renamed to cov()

  • Run script... is now in the Tools menu

0.7.3 (10/11/2019)

  • poisson() function for Poisson regression

  • negative numbers are now parsed correctly in String to Number conversion

  • fix split() method in String type

0.7.2 (09/11/2019)

  • lm() function for linear regression

  • logit() function for logistic regression

  • read_matrix() and write_matrix() functions to read/write a numeric array to/from a text file

  • DFT now uses double precision

  • slice() method to obtain a slice of an Array

0.7.1 (07/11/2019)

  • license is now GPL 3

  • Gaussian window for spectrograms and LPC analysis

  • improved LPC analysis

  • experimental automatic formant selection using Weenink’s method in formant queries

0.7.0 (05/11/2019)

  • formant queries (Analysis > Analyze formants...)

  • fix formant bandwidth estimation

  • maximum bandwidth parameter in formant analysis

  • fix path compression in project files

0.6.3 (02/11/2019)

  • fix regression in report_formants() due to the new array indexing syntax

  • documentation for the List type

  • anchors are now only edited on the visible layers

  • new statistical functions: covrc() (covariance) and corr() (Pearson’s correlation coefficient)

0.6.2 (31/10/2019)

This release brings improvements to the scripting engine, statistical functions, and fixes a regression.

  • new concatenation operator &

  • improved Array type

  • the multiplication and division operators can now operate on two arrays, or on an array and a scalar value

  • mathematical functions for numbers and arrays

  • statistical functions: sum, mean, variance, standard deviation and hypothesis testing (chi-squared test, F-test, one-sample t-test, two sample independent t-test with or without equal variance)

  • to_string() method for lists, arrays and objects

  • comment/uncomment selection in script views

  • bug fixes

0.6.1 (27/10/2019)

  • improved LPC analysis

  • scripting functions to convert between Hertz and bark, ERB units, mel and semitones (see Sounds). These functions accept a Number or an Array.

  • get_annotations() and get_sounds() now return a sorted list

  • automatic indentation in script views

0.6.0 (25/10/2019)

This release brings more sound visualization and analysis options, as well as a number of enhancements and bug fixes.

  • spectrogram in sound and annotation views

  • LPC-based formant tracking

  • waveform scaling using global, local or fixed magnitude

  • intensity settings

  • click on middle button (wheel) to zoom on the active selection

  • user dialogs

  • uninstall plugin (Tools > uninstall plugin)

  • new resampler

  • resample and/or convert sound to WAV, AIFF or FLAC

  • Sound objects are now accessible from the scripting engine

  • measure pitch, intensity and formants under the cursor

  • show/hide layers in annotation views

  • Export annotation(s) to plain text... (in File > Export)

  • updated documentation

0.5.2 (04/10/2019)

  • new import dialog for metadata

  • bug fixes

0.5.1 (29/09/2019)

  • new regular expression engine based on PCRE2

  • faster loading time for TextGrid annotations (~ 23%) thanks to the new regex engine

0.5.0 (27/09/2019)

  • new website at http://www.phonometrica-ling.org

  • create and edit annotations

  • scripting API to access the content of annotations

  • export metadata to CSV

  • bookmarks

  • fix initialization on Windows when the user’s directory contains non-ASCII characters

0.4.1 (21/09/2019)

  • fix communication with Praat on Windows when the user’s directory contains non-ASCII characters

  • better monospace font on Windows

  • improved preferences dialog

0.4.0 (20/09/2019)

This is the first functional version of Phonometrica. It brings the following features:

  • project management

  • native format for annotations based on annotation graphs

  • conversion between Praat TextGrids and Phonometrica annotations

  • typed properties (Boolean, numeric or textual)

  • query editor for single layer queries

  • query protocols

  • plugins

0.3.0 (30/08/2019)

  • initial implementation of annotation views

0.2.0 (17/03/2019)

  • project management, with support for metadata

  • script editor and scripting console

  • basic interaction with Praat

  • initial documentation

  • installers for Windows, macOS and Linux (Debian/Ubuntu)

0.1.0 (26/02/2019)

  • Scripting engine based on MuJS 1.0.5.

Phonometrica is partly based on Dolmen, developed and maintained by Julien Eychenne from 2010 to 2018. A python proof-of-concept of Dolmen was sketched out in April/May 2010. Dolmen was a complete redesign of the PFC platform (2006/2008), a concordancer implemented in Python and specifically written for the PFC project (www.projet-pfc.net).