Vowel Lab

Vowel chart

trail English reference vowels *Gabriel's vowels

presets English references (P&B) *Gabriel's vowels My sounds current sound

Drag the dot; sound plays while you hold it. Up = higher tongue (lower F1). Left = tongue further front (higher F2). Grey rings are Peterson & Barney (1952) American English averages; click one to hear it.

Spectrum

harmonics (from the source) formant envelope (from the filter)

Spectrogram

The source-filter idea: the spikes are harmonics of F0, made by the vocal folds. The smooth curve is computed from the formant settings. Change F0 and the spikes move while the curve stays put; change a formant and the curve moves under the spikes.

Vowels and controls

*Gabriel's vowels

Hold sound on Source

Move a slider or drag the dot, and this line will say what changed.

My sounds

Saved sounds keep F0, formants, bandwidths and source type. They are stored in this browser only, and also appear on the chart (◆) and in the continuum and oddball menus.

Readout

	Hz	Bark	ERB-rate	mel

Highlighted when two formants are within 3 Bark of each other, about the distance at which the ear merges two peaks into one (Syrdal & Gopal 1986). That convergence is what makes a vowel focal. Focalization is graded, not on/off.

Pick a tour. Each takes under a minute and moves the controls for you while you listen.

Vowel A Vowel B Steps 7 Spacing

Gap between onsets 800 ms

A continuum steps from A to B. Each step changes F1, F2 and F3 together; F0 and bandwidths come from the current sliders. Steps also appear on the vowel chart and can be picked as oddball standard or deviant.

Step	F1	F2	F3	ΔF1 / ΔF2 / ΔF3 (Hz)	Δ Bark

Standard Deviant Deviant probability 15% Trials 80 Onset-to-onset

500 ms Sound duration 250 ms no two deviants in a row

Choose a standard and a deviant, then start a block.

Blocks you've played

None yet.

Why reverse roles? In an oddball study the brain's MMN to "A standard, B deviant" is compared with "B standard, A deviant". If one direction gives a bigger response, that is a directional asymmetry. The Natural Referent Vowel framework (Polka & Bohn 2011) predicts that a change from a less focal vowel toward a more focal one is easier to notice. Listen to both directions and judge for yourself. The short interval is typical of MMN work; the long one (about 1600 ms) is what Polka et al. (2021) used to push listeners toward "phonetic mode" processing.

Teaching demo only. Browser audio timing is not precise enough for EEG. Do not use this to present experimental stimuli.

Drop a recording (.wav, .mp3, .m4a, ...) anywhere on this tab and it is analysed right here in your browser, the way Praat's To Formant (burg) does it. Nothing is uploaded. Or paste a Praat formant table (in Praat: select the Formant object, then Tabulate → List..., or the Formant listing from the editor), or import a .txt / .tsv / .csv file. A header row like time(s) nformants F1(Hz) B1(Hz) ... is needed. --undefined-- cells are skipped.

Import table or audio

From audio ⓘceiling Hz · formants · skip quiet frames Bandwidth limit 300 Hz Wide formants

Weighted mean

Selection to s

Drag across the plot to choose a stretch; double-click selects everything. With a recording, the shaded background is its wideband spectrogram (stronger shading = louder), so you can check the dots sit on the formant bands. Big solid dots have narrow bandwidths (well-defined peaks). Small hollow dots were excluded by the bandwidth limit. Grey = F4/F5, shown for context only.

	n used	Median	Weighted mean	Mean ± SD	IQR	Median B
No data yet.

"Use" loads the F1–F3 values and median bandwidths into the synthesizer and plays them; F0 stays as set on the sliders. Remember this is a resynthesis sketch of the measurements, not the original recording.

Advanced settings

extra formants for the synthesizer

Formants in the synthesizer

What F4 and F5 do. F1–F3 decide which vowel you hear. F4 and F5 sit higher (around 3500 and 4500 Hz for a man, a few hundred Hz higher for a woman) and barely change vowel identity. They change the voice: adding them makes the sound brighter and less "buzzy", closer to a real talker, and they carry some of what makes one speaker sound different from another.

They are added as two more resonators in the same cascade, which costs almost nothing to compute. Loudness is re-normalized, so switching them on changes the timbre, not the volume. F4 is kept at least 150 Hz above F3, and F5 150 Hz above F4. Switching between male and female voice resets them to approximate defaults for that voice.

Sounds you save remember their F4 and F5. With extra formants on, Use medians in the Analyze tab also loads the measured F4 and F5 (when the table or recording has them).