39 sounds · 126,036 words · offline

See a word pronounced.

Hearing a word tells you what it sounds like. It does not tell you what to do with your mouth. LinguaKata moves an avatar's lips, tongue and jaw through the word — slowly enough to copy — and shows a cross-section of where the tongue goes and where the air comes out.

Free. No account, no microphone, nothing to install.

Front
Cross-section

air contact point voicing

Built on the CMU Pronouncing Dictionary. Runs in the browser and on iPhone.

0.1x playback Step by sound IPA + respelling Stress & reduction Six avatars
What you see

Six things a recording cannot show you

Each of these is modelled from phonetics, not animated by feel — and each one is a mistake a learner makes because nothing ever showed it to them.

Where the tongue is

Tip or body, front or back, touching or merely close. /s/ has the tip near the ridge — press it all the way up and /s/ becomes /t/.

Where the air goes

Out of the mouth, out of the nose, or nowhere at all while pressure builds behind a closure. The line freezes when you hold a sound, because air streaming through a held stop would be a lie.

The puff, and when there isn't one

The /p/ of "pin" has a puff of air. The /p/ of "spin" does not — after /s/ English drops it. This is mostly why a Spanish or French "pin" is heard as "bin".

A

Which syllable the word leans on

PHO-to-graph against pho-TOG-ra-phy: one morpheme, different beat. Put the stress in the wrong place and a correct word stops being understood.

ə

The vowels that go vague

Nearly half the vowels in English are unstressed, and they undershoot. The avatar's weak syllables look weak, because copying a fully articulated schwa teaches the wrong thing.

Whether the voice is on

And when it isn't: English /b d g/ at the start of a word are usually voiceless too. The larynx lights up only when it should.

The argument for the cross-section

These two mouths are identical

/p/ and /m/ have exactly the same lips — not similar, the same. Both pictures below come from the app's own drawing code, so this is not a figure drawn to make a point; it is the point.

Notation

Three ways to write it down, at once

IPA is precise and unfamiliar. A newspaper respelling is familiar and throws information away. So the app shows both, and highlights the exact part being spoken as it plays.

The app also says what each system loses. A respelling that writes both "thin" and "this" with th has quietly merged two different sounds, and a learner who trusts it will merge them too.

Honest limits

What it does not do

Worth knowing before you decide it is the tool you wanted.

It does not listen to you

No microphone, no recording, no score. It shows you the movement and you copy it. Comparing your formants to a model is a genuinely different app and a much bigger one.

The timing is modelled, not measured

Phones report when speech starts and stops and nothing in between, so durations come from a model of English and the mouth can drift from the audio across a long word. Below 1x the voice finishes first — the app says so rather than pretending otherwise.

General American, one accent

The dictionary behind it is American English. A British or Australian learner will find the vowels of some words are not the ones they say.

A guess is labelled a guess

Words outside the dictionary are worked out from spelling rules and badged as an estimate. English spelling being what it is, a guess presented as fact is the one genuinely harmful thing this app could do.

Free

Type a word. Watch it.

Runs in this browser with nothing to install. Judge the shapes on the web and the audio on a phone — browser voices differ from the ones your device ships.

iPhone build in TestFlight; Android is not built yet. Questions to support@spaikz.com.