Learn Thai by speaking it
Thai is tonal, and a tone is not an accent detail. Say the same syllable with a different pitch contour and it is a different word, which is why a text transcript is nearly useless for correcting you.
Last reviewed
The short answer
Thai is the strongest case in this entire app for a tutor that hears your voice. Tone is phonemic, so pitch carries meaning rather than emphasis, and a transcript records the word it thinks you meant while throwing away the pitch contour that decided it. An app working from text literally cannot correct your tones. This one receives the audio itself. Thai script is shown with a Latin reading beneath so a beginner is never locked out.
Thai is the language in this app where the case for hearing the learner is not an argument. It is arithmetic.
Why a text-based app cannot teach you Thai
Thai has five lexical tones. The pitch contour on a syllable is not emphasis or attitude, it is part of the word, in the same way a consonant is.
Now consider what a speech-recognition system does. It takes your audio, decides which word you most likely meant, and outputs that word as text. The pitch contour that determined the answer is discarded at that moment, because text does not encode it.
So an app that transcribes you and sends the text to a language model has already thrown away the only information that could tell it your tones were wrong. It can correct your grammar and your vocabulary. It structurally cannot correct the thing Thai learners most need corrected.
Here the audio goes to the model directly and comes back as audio. Nothing transcribes you first.
What that changes in practice
The tutor can hear that you said the right word with the wrong contour, and do what a teacher does: say it back correctly inside its own reply, in the flow of the conversation, so you hear the shape in place.
It also means the thing you are practising against is a real speaking rhythm rather than a series of isolated words. Tones are considerably harder in connected speech than in a drill, because English speakers unconsciously import sentence intonation, and that is only visible when you are actually talking.
Script, and the reading underneath
The tutor writes in Thai script. A Latin reading of your own speech is shown underneath by default, which matters more here than in most languages: Thai script does not separate words, and the tone rules interact with consonant classes in a way that takes a while to read fluently.
That default is deliberate. Somebody who can already read Thai switches it off in a moment; somebody who cannot has no way to discover the setting exists.
What a session is
Unscripted conversation with something you can interrupt, backed by an authored CEFR syllabus of 333 items from A1.1 to C1, of which 114 are capabilities: concrete things you can do rather than grammar points. The pronunciation patterns above are among 20 authored for Thai specifically, most of which are about tone.
The five tutors span the country, which matters in Thailand more than in most places. Bangkok is the standard, Chiang Mai is gentler, and Isan and the south will stretch your ear considerably.
Going to Thailand
Travel Mode changes the app for the length of a trip: rehearsal before you go, one real-world mission a day while you are there, set in the city you are in and pitched at or below your level, and a debrief with your tutor afterwards.
Thailand rewards this unusually well. Effort in Thai is met with genuine delight, the everyday transactions are formulaic enough to rehearse, and the politeness particles are the kind of thing that transforms how you are received and takes about a minute to learn.
The five Thai tutors
Every language has five, each from a different place and each sounding like it. You can switch whenever you like, and the app remembers where you are regardless of who you talk to.
NuchBangkok
Central Thai, the standard form, with clear five-tone distinctions.
SomChiang Mai
Central Thai spoken with the soft, unhurried warmth and gentle politeness of the north.
PraewChonburi
Eastern Thai from the seaboard, bright and quick with the easy, open warmth of the coast.
TanKhon Kaen
Northeastern Thai from the Isan heartland, with the rising, sing-song lilt and warmer vowels that set Isan speech apart from Bangkok.
PloyHat Yai
Southern Thai from Songkhla, with the fast, clipped rhythm and shortened tones the south is known for.
What English speakers usually find hard
These are drawn from the pronunciation patterns the tutor listens for in Thai. Because it hears your voice rather than a transcript of it, they are things it can actually notice and work on with you.
- Five tones, and they are the word
- Mid, low, falling, high and rising. The same syllable said five ways is five different words: maa can be come, horse or dog depending on the contour. This is not emphasis and it is not optional, and it is the first thing to get right rather than a refinement to add later.
- English intonation actively destroys Thai tones
- The most common English-speaker error is not producing tones badly, it is producing English intonation on top of them. English raises the pitch at the end of a question regardless of the words, which converts low, mid and falling tones into rising ones. Thai marks questions with a particle instead, so the contour has to stay put.
- Vowel length changes the word too
- Short and long vowels are both contrastive, independently of tone. Khâao and khâo are rice and enter. Thuk and thúuk are every and cheap. English compresses everything toward a neutral length, and getting this right is the second-biggest win after tones.
- Final stops are closed and never released
- A final k, p or t is formed and then simply held. English releases final stops with a small puff of air, and doing that in Thai sounds immediately foreign. The mouth closes and stays closed until the next syllable.
- There are no spaces between words
- Written Thai runs words together and marks phrase boundaries with spacing instead. Working out where one word ends and the next begins is a skill built by ear as much as by eye, which is another reason listening comes first here.