Can ChatGPT or Gemini check my pronunciation — or my Chinese tones?

There are three different situations, and they are easy to mix up.

• Typed chat: the assistant receives only the text you type. It can explain pinyin, tone rules or English vowels, write practice sentences and tell you how a word should sound — but there is no recording to judge.
• Dictation (the microphone button that turns speech into text): your speech is converted into the nearest real words before the model sees it. A slightly wrong tone or vowel is often written down as the word you meant, so the mistake disappears.
• Voice mode (ChatGPT’s voice conversations, Gemini Live): the assistant takes in your audio and can say whether something sounded off. What it says is generated feedback, not a measured score per sound or syllable, and it can differ between tries, versions and apps.

We have not published an accuracy comparison, so this page makes no claim about which tool is more accurate. You can test it yourself:

• Go somewhere quiet. Pick a word whose tone or sound you know — 马 mǎ (horse), or English “ship”.
• Say it correctly and ask for feedback. Then say it wrong on purpose: mǎ as mà (a sharp fall), or “ship” as “sheep”.
• A tool that is really listening must flag the second take — and a good one says which syllable and which tone or sound was wrong.
• Repeat each pair three times. One take can be off; the pattern over three takes is what counts.

Assistants are good at explaining rules, making example sentences at your level and answering “why”. A scorer is good at measuring the same thing the same way every day, so you can see whether a sound is improving. Many learners ask the assistant for sentences, then record them in a scorer.

TonePerfect’s limits: background noise, whispering, very creaky third tones, neutral tones and very fast speech make any automatic check less reliable. It scores what you read or repeat, not a free conversation, and it gives no exam level.