Transcribe audio and video for free
No account, no daily limit, no upload. Your file stays on your device.
How it works
The first time you use it, your browser downloads a speech model (165 to 560 MB) from Hugging Face and keeps it. After that, your computer does the work. Your audio is read and transcribed inside this browser tab and is not sent to Kingy or anyone else.
What we measured
We ran real US Supreme Court arguments through it and compared the results with the official transcripts. Tests ran November 1, 2026 on a MacBook Pro with an M4 Pro chip and 24 GB of memory.
| Test | Result |
|---|---|
| 2 hr 26 min argument, Best model | 9.0% word error rate, finished in about 10 minutes |
| Same file with speaker labels | 12 real speakers became 13 labels, 98.1% of words under the right label, about 21 minutes |
| 5 minutes with heavy added noise | 14.4% word error rate |
| Older-computer option limited to 2 threads | About 1.6 times faster than real time |
Court transcripts record every false start, so part of that error rate is the model tidying up speech. Your results will depend on your audio and your computer. On a Windows gaming laptop using its built-in Intel graphics, not a separate graphics card, Chrome finished the same 2 hr 26 min file in about 30 minutes.
Questions
Is there really no limit?
We set no limit on files, length or number of uses, because your computer does the work and we pay nothing per file. The practical limit is your computer's speed and memory.
Does my audio get uploaded?
No. The page never sends your file anywhere. The only download is the speech model itself. Hugging Face, which hosts the model, sees an ordinary request for that file. The page counts how many transcriptions start, finish or fail each day, with nothing else attached: no file name, audio, text or anything that identifies you.
Which browsers and devices does it work on?
We tested Chrome and Edge on a Mac and Chrome on a Windows laptop. Safari works but ran slower in our tests, so it starts on Fast. Firefox has not been tested yet. It is made for laptops and desktop computers. Phones and tablets are not supported yet: it did not work on the iPhone we tried, so use a computer for now.
Can I use other tabs while it runs?
It keeps going, but keep this tab in front if you can. With speaker labels on, a 5-minute test ran about six times slower when another tab was in front. Opening a separate browser window for other work may help, but we have not measured that.
What if I close the tab?
Progress is saved in this browser. Choose the same file again and it carries on from where it stopped.
How accurate are the speaker labels?
They were very accurate on clear recordings of people taking turns. Expect more mistakes when people talk over each other or sound alike. Labels are numbered; click one to type a name.
Should I rely on this for legal or medical records?
Not without checking it. Speech models make mistakes, and sometimes write something plausible that nobody said. Check anything important against the recording. Click a timestamp to hear that part.