Vowel Formant Filter (a-e-i-o-u) — User Guide

Shapes a sound with Praat's built-in vocal-tract models for a, e, i, o, u. Each selected vowel produces a complete filtered version of the source; the versions can optionally be concatenated into one sequence.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.5 (2026) License: MIT License Repo: Praat AudioTools
Contents:

What this does

The script applies a fixed vowel-shaped vocal-tract filter to the source sound. For every selected vowel, Praat creates its built-in VocalTract model, converts that model to an LPC filter, and filters the complete source through it.

One vowel = one complete version of the source. Selecting several vowels does not morph continuously between vowels inside the sound. The script renders separate full-length versions and, when Concatenate_results is enabled, joins them in the fixed order a → e → i → o → u, omitting any vowels that were not selected.

The vowel character comes from the resonances of Praat's vocal-tract area-function models. The source supplies the excitation and musical material; the LPC filter reshapes its spectrum toward the selected vowel.

Channel behavior: processing is performed on a mono copy. Mono input remains mono; stereo or multichannel input is converted to mono before filtering, so the original spatial image is not preserved in the processed outputs. The original Sound object itself is not modified.

Quick start

  1. In Praat, select exactly one Sound object.
  2. Run script… → a-e-i-o_filter.praat.
  3. Choose a Preset. The default is All Vowels (a-e-i-o-u).
  4. Use Manual only when you want to choose the vowel checkboxes yourself.
  5. Choose whether to concatenate the rendered vowels, draw the visualization, and play the final result.
  6. Click OK.
Default result: five full-length mono versions are rendered for a, e, i, o and u, then directly concatenated into one output named <source>_VT_AllVowels. Its duration is five times the source duration.

Presets & Manual

Presets 1–9 determine the vowel selection and override the five Include_* checkboxes. The checkboxes are read directly only when Manual is selected.

PresetRendered vowelsOrder
All Vowelsa, e, i, o, ua → e → i → o → u
Classica, e, i, oa → e → i → o
Front Vowelse, ie → i
Back Vowelsa, o, ua → o → u
Single: aaa
Single: eee
Single: iii
Single: ooo
Single: uuu
ManualAny combination of a/e/i/o/uAlways follows a → e → i → o → u
Manual: at least one vowel must be enabled. If all five checkboxes are off, the script stops with an error message.

Controls

ControlDefaultBehavior
PresetAll VowelsSelects one of nine fixed vowel sets or Manual.
Include_aOnUsed only by Manual. Choose which vowel filters are rendered.
Include_eOn
Include_iOn
Include_oOn
Include_uOff
Concatenate_resultsOnWhen more than one vowel is rendered, joins the separate versions directly into one Sound.
Draw_visualizationOnDraws spectral diagnostics: vowel filter responses, input/output spectra, F1/F2 vowel space, formant values and per-vowel output peaks.
Play_resultOnPlays the final selected output after processing.

How it works

The processing path is:

Selected Sound → mono processing copy → Praat vowel VocalTract model → VocalTractTier → LPC filter → vowel-filtered Sound → optional concatenation
  1. Mono processing source. A mono copy is made for LPC filtering. Multichannel input is mixed down first.
  2. Vocal-tract target. For each selected vowel, the script calls Praat's built-in phone model for a, e, i, o or u.
  3. Tier and LPC. The vocal-tract model is inserted at the midpoint of a VocalTractTier spanning the source duration, then converted to LPC with a 5 ms time step.
  4. Filtering. The mono source is filtered by that LPC with Use LPC gain = no.
  5. Sample rate. Each filtered result is resampled to the original sampling frequency using Praat resampling precision 50.
  6. Combination. If requested and more than one vowel exists, the vowel versions are concatenated directly in canonical vowel order. There is no overlap, crossfade or added gap.
Level: the script does not normalize or apply a peak ceiling. LPC filtering can substantially change amplitude. Version 0.5 reports the final peak in the Info window and warns when it exceeds 1.0; this is reporting only and does not change the samples.

Outputs

Individual vowel names

Each rendered vowel is initially named:

When Concatenate_results is on

If two or more vowels are rendered, the individual vowel objects are joined and then removed. The retained result is named <source>_VT_<preset>, for example <source>_VT_AllVowels, <source>_VT_Classic or <source>_VT_Manual.

The concatenated duration is N × source duration, where N is the number of selected vowels. A single-vowel preset remains a single <source>_VT_vowel object even when concatenation is enabled.

When Concatenate_results is off

All rendered vowel objects remain in the Objects list. The script uses the first selected vowel in a → e → i → o → u order as finalOutput; that is the object selected at completion and the one played when Play_result is enabled.

PropertyProcessed output
ChannelsMono
Sample rateRestored to the source sampling frequency
Per-vowel durationSource duration
NormalizationNone

Visualization

Version 0.5 replaces the earlier stacked waveforms with a spectral view designed around what the vowel filters actually change. The Picture window uses a fixed 8 × 9.2 inch page and contains four diagnostic areas plus a summary strip.

1. Vowel filter responses

The first panel plots the response of the actual LPC object used for each selected vowel. Frequency is shown on a logarithmic axis. Each vowel curve is normalized to its own maximum, so this panel compares resonance shape, not absolute gain.

2. Input and filtered spectra

The second panel compares the mono processing input in gray with every filtered vowel output on one shared dB scale. This makes both spectral tilt and absolute level differences visible. Unlike the old waveform display, a high-gain vowel can no longer look identical in height to a low-gain vowel merely because each panel was auto-scaled separately.

3. F1/F2 vowel space

The lower-left panel places the selected vowels in F1/F2 space using the usual phonetic orientation: F2 decreases from left to right and F1 decreases upward. The points come from Praat's LPC → Formant conversion at the midpoint of the source, not from peak-picking the drawn spectrum.

4. Formants and output level

The lower-right table reports F1, F2, F3 and the absolute output peak for each rendered vowel. A peak above 1.0 is shown in red with an exclamation mark, and the panel adds a clipping warning.

Important: this warning does not apply gain reduction. Version 0.5 deliberately leaves the audio level unchanged from v0.4.3. If a peak exceeds 1.0, scale the output yourself before saving to an integer PCM format.

Summary strip

The bottom strip reports the preset, selected vowels, source name, source duration and sample rate, output arrangement, VocalTractTier insertion time, and the frequency span used by the visualization. The analysis range starts at 80 Hz and extends to the lower of 8 kHz or approximately 98% of Nyquist.

What the plots mean: panel 1 compares the shapes of the vowel filters themselves; panel 2 shows what those filters did to this particular source on a shared level scale; panel 3 shows the vowel locations from F1/F2; panel 4 gives the numeric formants and output peaks. The visualization is diagnostic only and does not alter the sound.

Further Reading