Hear your own Chinese tones

Pick a tone, listen to the model, then press Record and say the syllable once.

Almost every free tone exercise — including the three on this site — plays you a syllable and asks which tone it was. That trains your ear, and your ear is usually not what is holding you back. The gap most learners are stuck in is between recognising a 2nd tone and producing one, and you cannot close it by listening, because you cannot hear your own pitch the way a listener does while you are busy producing it.

What closes it is seeing the shape. A Mandarin tone is a movement of pitch across a syllable, and once your movement and the target movement are drawn on the same axes, the mistake is usually obvious in one look: the 2nd tone that starts high and has nowhere left to rise, the 4th tone that trails off instead of dropping, the 3rd tone said as a low flat note with no dip in it at all.

The neutral tone is deliberately not drilled here. It has no shape of its own — it is short, quiet and takes its pitch from the syllable before it — so there is nothing to draw a target for. The Tone Trainer covers it by ear.

It stays in your browser. The page measures the pitch of the recording — how fast your vocal folds are vibrating, moment by moment — which is arithmetic over the raw sound and needs no server and no knowledge of what you said. There is no speech-recognition service involved, nothing is uploaded, no file is created, and nothing is written to storage: each take replaces the previous one in memory and the tab closing is the end of it.

The microphone is released the moment a take stops, so your browser's recording indicator goes out between attempts rather than staying lit while you read the page. See the privacy policy for what this site does and does not collect elsewhere.

Both lines are drawn in semitones relative to your own average pitch in that take, never in raw frequency. That is not a simplification, it is what a tone is: Mandarin speakers with low and high voices produce identical tones in completely different frequency ranges, and a chart in hertz would show two people saying the same thing as two different answers.

The score is the agreement between the two shapes, and deliberately not how closely you matched the size of the model's movement. The dashed target is this site's teaching contour — the same exaggerated shape the Tone Trainer plays, drawn more dramatically than natural speech so a beginner can hear it. Marking you down for producing a natural-sized 2nd tone would be the tool grading you against its own exaggeration. So the shape is what is scored, and the size of your movement is reported underneath as information.

If a take cannot be read, the page says so instead of drawing a line. A pitch detector fed a whisper, a cough or room noise still produces numbers, and a drawn curve looks equally confident whether or not there was a voice behind it — so a take that is too quiet, too short or unvoiced is reported as unreadable rather than guessed at.

Every syllable here is a tone minimal set: all four tones of it are four different words. That is the argument for practising tones at all — mā, má, mǎ and mà are not four ways of saying one thing, they are mother, hemp, horse and to scold. Syllables that are missing a tone are left out rather than padded with something invented, which is why shǔi and liáng are absent: there is no first-tone shui to practise.

Readings and meanings come from the same dictionary the translators use, resolved when the page is built rather than written out beside each character — so this table cannot end up disagreeing with the rest of the site about a word. A character the dictionary does not confirm in that tone is left out rather than shown.

• Listen before every attempt, not just the first. The model resets your ear; without it you drift toward whatever you said last time.
• Say the syllable on its own and slightly long. A tone needs time to move. Rushing is the single most common reason a shape comes out flat.
• Work down one syllable's four tones in order, then jump between two of them at random. Producing a tone in isolation is easier than switching, and switching is what speech actually requires.
• Use "Hear my take". Reading the chart tells you what happened; hearing it back is what makes the correction stick.
• Do not chase 100%. Anything above roughly 80% is the right shape. Past that you are matching an exaggerated teaching contour, which is not what you want to sound like.

Is my voice sent anywhere?

No. The recording is measured inside your browser and never uploaded — not to this site and not to anyone else. No speech-recognition service is involved, because measuring pitch does not require knowing what was said. Nothing is stored: each take replaces the one before it in memory, and closing the tab ends it.

Why does it say it could not read my take?

Usually it was too quiet, whispered, or started before the recording did. A whisper genuinely has no pitch to measure. Speak at a normal volume, press Record first and then say the syllable, and keep the take to a single syllable rather than a sentence.

Does a deep or high voice affect the score?

No. Everything is drawn in semitones relative to your own average pitch in that take, so only the shape of the movement is compared. The frequency your voice sat at is shown underneath purely as information.

My 3rd tone looks nothing like what I hear in conversation.

That is expected. Alone, the 3rd tone is the full dip: down, low, back up. In a sentence it usually stays low without the rise, and before another 3rd tone it turns into a 2nd tone entirely — the change covered on tone rules and sandhi. The full shape is the one worth drilling in isolation.

Can I use this on a phone?

Yes, and it is the better device for it — a phone's microphone is close to your mouth, which is exactly what a pitch reading wants. Your browser asks once for microphone permission and you can withdraw it at any time from the address bar.

Does it work for words, not just single syllables?

Not yet — this page drills one syllable at a time, because that is where a tone has a target shape you can compare against. For whole words and sentences, the speaking practice game checks whether what you said was understood, and the text reader will read any passage aloud at a speed you choose so you can shadow it.

• Pinyin Tone Trainer — the ear half: hear a tone and identify it, with exaggerated audio
• Tone Pairs Quiz — the pairs that actually cause trouble, drilled two at a time
• Tone Pair Arcade — the same skill against a clock
• Interactive Pinyin Chart — every Mandarin syllable, with audio for all of its tones
• Chinese Tone Rules and Sandhi — why a tone changes when it meets another one
• Speaking Practice — say a whole word and see whether it was understood
• Text-to-Speech Reader — paste any Chinese text and shadow it at your own speed