You speak
Talk normally into a laptop or phone. No special microphone and no training session first.
For people with unclear speech
Say it once.
Be understood.
Illness and surgery can make speech hard for other people to follow, and almost impossible for phones and computers to follow. OwnVoice listens and turns Mandarin speech into text that anyone can read.
Free. Open source. Your voice never leaves your computer.
What it does
It listens
A speech model built for unclear speech turns what you said into Mandarin text, a phrase at a time.
They read
The person you are talking to reads it on the screen, or on a projector in a bigger room.
Hear the difference
This is one sentence from a real recording. Use the buttons to hear it get harder to understand, then read what ordinary speech recognition makes of it and what OwnVoice makes of it.
與地主做良性的溝通
“to communicate constructively with the landowner”
As it gets harder, the pitch flattens and the words stretch out. In Mandarin the pitch is what tells two words apart, so that is exactly the part that carries the meaning.
Ordinary speech recognition
44% wrong
與地主做人心的構同。
The key word is lost. 溝通, meaning communicate, became 構同, which is not a word.
OwnVoice
11% wrong
與地主做人性的溝通
The sentence is readable. One word is still wrong, so this is a real improvement rather than a perfect one.
Try it with your own recording
Open the live demo, upload a Mandarin recording and get the OwnVoice transcript. It runs our model online on a free server, so it can take up to a minute. With the model running on your own computer, the same page also captions you live and simulates dysarthria on your own voice.
Try it with your own voice
Press start, allow the microphone, and speak Mandarin. Pause between sentences. Each pause ends a phrase and the text appears below.
Your voice is sent to the model, turned into text, and discarded. It is never stored.
Where it helps
The same tool, used in very ordinary places.
At home
Families stop guessing. Instead of asking someone to repeat themselves four times, the sentence is simply on the screen.
At a medical appointment
A patient can describe their own symptoms instead of relying on a relative to speak for them.
On video calls
The browser extension captions whatever is playing in a tab, so an online consultation or a work meeting is readable live.
At a service counter
Banks, clinics and government offices can read a request rather than turning someone away.
In a classroom or a hall
Full screen mode puts the captions on a projector so a whole room can follow a speaker.
For speech therapy
Therapists get a written record of what was understood, which is hard to judge by ear alone.
How to use it
-
1
Open it
Use the website in a browser, or install the extension if you want to caption video calls and other tabs.
-
2
Press start and talk
Allow the microphone once. Speak normally and pause between sentences. Each pause ends a phrase and sends it to be read.
-
3
Show the screen
Hand over the phone, turn the laptop around, or switch to full screen for a projector. Change the interface to Chinese whenever you like.
How well it works
Plain numbers, measured on speakers the system had never heard before.
41%
fewer mistakes on the hardest speech, compared with ordinary speech recognition
86%
of characters read correctly overall, up from 81%
Under 1s
to caption a second of speech, so captions keep up with talking
Being straight with you. These results were measured on recordings we degraded on purpose to imitate unclear speech, not yet on recordings from patients. It works, and we are honest that clinical testing is the next step rather than something already finished. Accuracy also varies from person to person, so try it before relying on it.
The Chrome extension
The website captions your own microphone. The extension also captions the audio of any browser tab, so a video call, an online class or a recorded meeting becomes readable while it plays. A web page cannot do that.
Install it
It is not in the Chrome Web Store yet, so it installs from a folder. That takes a minute and is completely normal for an extension you build yourself.
- Download the extension below and unzip it. You get a folder called
ownvoice-extension. - Open
chrome://extensionsand turn on Developer mode, top right. - Click Load unpacked and choose the
ownvoice-extensionfolder. - Open the tab you want captioned, then click the OwnVoice icon on that tab.
- Pick Microphone or This tab’s audio, then press start.
Two capture modes
Your microphone, or the sound coming out of the tab you are on.
English or Chinese
Switch the whole panel between English and 中文 at any time.
It needs the model running
The extension shows the captions but does not contain the model, so the recogniser has to be running somewhere. That can be your own computer.
Coming soon
The OwnVoice app
The extension covers calls on a computer. The app covers everywhere else: at work, at a service counter, at the doctor. Speak, and your words appear on your phone. One tap shows them in large text to the person in front of you.
Talk
Live captions in Traditional Chinese, with an English line under each one.
Show
One tap turns the last sentence into big text for a coworker, a pharmacist or a clerk.
Private by default
Audio is deleted after captioning. History stays on the phone. Sharing recordings for research is off unless you agree.
App Store · coming soon
Google Play · coming soon
We are building the app now. Until it ships, the website and the Chrome extension work today.
Your voice stays yours
The speech is processed on your own computer and then discarded. Nothing is uploaded, nothing is stored, and there is no account to create. That is also why a clinic can try it without a data protection review first.
Get started
OwnVoice is free and open source. Use it, read how it works, or build on it.
Mandarin today. Taiwanese Hokkien and Cantonese are next.