MirrorSay began as a tool I made for my own family.
We all wanted the same thing: to open our mouths and have English come out the way we meant it. We could read it. We could write it. But in the moment, the words came out in the wrong order, or didn't come out at all. None of the apps we tried quite fit, so I built the one we needed.
Today it's an English speaking app that anyone can use: say a sentence, see how a native speaker would say it, hear it read aloud, then shadow it with word-by-word rhythm feedback. In mainland China it's called 像话口语; in Japan, ミラーセイ.
Who's behind it
One independent developer. For now MirrorSay is a labor of love: the servers, the speech models and every AI rewrite are paid for out of my own pocket.
I'm going to keep maintaining and improving it.
Why it works the way it does
When I tried to speak, the problem was rarely that I didn't know enough English. It was that:
- what came out followed my native word order, and I could hear it sounded off, but had nobody to tell me how a native speaker would put it;
- there was no one to ask in the moment. Tutors need booking. AI chat apps expect you to keep a whole conversation going, which is its own kind of pressure;
- the textbook sentences I'd memorized never showed up when I needed them, because they were never things I wanted to say.
So MirrorSay does one small thing well: it mirrors the sentence you actually want to say back to you in a native speaker's words, then lets you say it a few times until it's yours. That's the name: Mirror + Say.
How it works
- Say a sentence. Anything you'd really say this week. Speech recognition runs on your phone whenever possible.
- See the native version. The recognized text is sent to my server, where a large language model rewrites it into natural spoken English and explains in one line what sounded off. Sentences that are correct but textbook-stiff get a more conversational version too.
- Listen, then shadow. Play the reference, slow it to 0.8× if you need to, then say it yourself. Your timing is compared word by word with the reference, and you can replay your own voice. It's a rhythm comparison, not a pronunciation score.
You can also practice a daily native speech pattern, import articles and bilingual subtitles, and save fixed sentences you want to get exactly right.
Your voice and your sentences
The privacy policy is the official version. In plain words:
- Your recordings stay on your phone. They're stored only on your device and deleted along with the sentence.
- Only text leaves your phone. Rewriting needs a connection, and only the recognized text is sent, never audio. The server returns the result and doesn't keep your sentences.
- No account needed. Signing in is optional and only syncs your saved sentences between devices.
- Analytics are anonymous. Things like "finished a shadowing round", never what you said.
Price
MirrorSay is free to start, with no ads and no sign-up. If I ever add paid features, I'll say so clearly and in advance.
The Speaking Guides
The Speaking Guides are written alongside the app, with a few rules:
- each one answers a question learners actually ask, and should get you speaking right away;
- they put what learners often say next to what native speakers say, with complete example sentences;
- research is cited only for what the source actually supports, with no invented numbers and no promised score gains;
- when MirrorSay comes up, only features that really exist are described.
If you find a mistake, tell me and I'll fix it.
Say hello
- Email: kawokala@gmail.com
- Or use the feedback option inside the app.
Bug reports, corrections, feature ideas, or just a note that it helped: all welcome.