Mandarin pronunciation practice, in your own voice
Mandarin’s four tones carry meaning the way vowels do in English — get one wrong and mǎi (买, to buy) becomes mài (卖, to sell). Most practice tools hand you a single reference voice, usually pitched well above or below where you actually speak, so you spend your attention transposing before you’ve even tried to copy the tone. Polly measures your own pitch first and moves the reference into it.
Why Mandarin tone practice usually fails
Mandarin has four tones plus a light, unstressed fifth ("neutral") tone: a high flat tone (mā, mother), a rising tone (má, hemp), a low dipping tone (mǎ, horse), and a sharp falling tone (mà, to scold). English uses pitch too, but only for sentence-level things like turning a statement into a question — never to change one word into a completely different word. That means English speakers have spent a lifetime training their ears to ignore exactly the signal Mandarin needs them to hear.
Add a second problem on top: almost every recorded reference voice, from textbooks to apps, sits in one fixed register. If your own speaking voice is a fifth below it, you’re not just learning the tone — you’re doing mental arithmetic to guess what that tone would sound like in your own voice, every single time you practice.
What changes when the reference is pitch-matched
Polly measures your median fundamental frequency from one recorded sentence, then resamples the Mandarin voice to that exact register — the shape of every tone contour stays identical, only its centre moves to where you actually speak. A rising second tone is still the identical rise; it’s just no longer a rise you have to relocate in your head before you can produce it.
After you repeat the sentence back, speech recognition checks what you actually said against the target, and a model scores accuracy word by word — with tone errors called out specifically, since a mis-heard tone is usually what turns a correct sentence into a different one to a native listener.