Flashcards teach you words. Books teach you Chinese.
This one listens while you read it out loud.
Point it at any Chinese e-book and it becomes a patient tutor. One sentence at
a time, large and centred, with pinyin printed above every
character. It reads the line aloud in a natural voice, then
listens to you say it back and colours every character green,
amber or red. Score 90% and it moves you on by itself. All of it runs on your
own computer, for free, and your voice never leaves the machine.
After you've spoken. Green is correct, amber is the right sound on the
wrong character, red is missed — and the score decides whether it advances.
3 colours
per character, not one score for the sentence. You find out
which character you missed.
90%
and it advances on its own, so a good run flows without touching the
mouse.
5 speeds
from 0.5× while a phrase is new up to 1.5× once it clicks.
Private
Speech recognition runs locally. No account, no subscription, no audio
upload.
Start · the gap it fills
Reading practice that can hear you
Most Chinese reading tools show you the text and, at best, read it to you.
The half that is missing is the half that matters for speaking: nobody checks
whether what came out of your mouth was right.
A tutor does that, and a tutor is expensive and only available on Tuesdays.
This does the narrow, mechanical part of the job — did you say the characters
on the screen — for free, on demand, at whatever hour you happen to be reading.
↓ the loop
Read, listen, speak, get corrected
Pinyin above every character
Tone-marked, sitting directly above the character, so you never break
the sentence to look something up.
One sentence at a time
The current line is large and centred; neighbours sit faded above and
below, like scrolling lyrics.
Sensible sentence sizing
Long paragraphs split at natural pauses, and lines that come out too
short get stacked back together. Each unit is a comfortable mouthful.
Simplified or Traditional
Flip the whole text either way from a menu, any time, whichever the
book was written in.
Natural voices, four ways
Male or female, Mainland or Taiwan accent. Once a line has been read,
replaying it is instant.
A gentler mode
Lenient grading counts near-misses that sound right, so
homophones stop feeling like a punishment.
Before you speak. Every button shows its keyboard shortcut.
↓ the marking
Amber is the interesting colour
Green and red are obvious. Amber — right sound, wrong
character — is the one that teaches you something, because it separates
"you can't pronounce this" from "you pronounced a different word
that sounds identical." Those are completely different problems and most
tools score them the same.
That distinction is also why Lenient mode exists. Mandarin has an enormous
number of homophones, and a learner who says the right syllable with the right
tone has done the thing being practised, even when the recogniser picks a
different character.
↓ staying out of the way
It remembers where you were
Close it and reopen it and you are back on the same book, chapter and
sentence, with your voice, speed and settings intact. Chapters are listed down
the side, with cover, copyright and contents pages flagged so you can skip to
the actual reading.
There are two interfaces doing the same job: a modern media-player layout
that behaves like a cross between a music app and a karaoke machine, and a
classic reader. Both have light and dark themes and a full set of keyboard
shortcuts, so the whole loop works without reaching for the mouse.
↓ what it can't do
It checks characters, not tones directly
The scoring works by comparing what the speech recogniser heard against the
characters on screen. That is a good proxy for pronunciation and it is not the
same thing as tone assessment — a recogniser can be forgiving in ways a native
listener would not be, and it can also mishear you when you were right.
It needs the internet the first time a sentence is spoken, to fetch that
line's audio. Everything afterwards, including the speech checking, runs on
your machine. And you supply your own books: the repository ships code only,
no copyrighted text.
Finish · what it is built on
Runs on your machine, costs nothing
Speech
Neural Mandarin voices, male or female, Mainland or Taiwan
Recognition
Runs locally. Microphone audio is never uploaded
Scoring
Per character: correct, right-sound, or missed, plus an overall accuracy score
Auto-advance
At 90% or higher, switchable off
Interfaces
A modern player and a classic reader, both light and dark
Books
You supply your own. Code only in the repository
Sibling
Cantonese Learner V2, the same loop applied to daily Hong Kong news and conversations
