Free Tool · No Upload · Auto language · Runs in your browser
Free Audio to Text
Turn recordings, lectures and voice memos into text. The Whisper AI runs entirely in your browser, so your audio never uploads to a server: private by design, and free to start. It detects the language automatically and gives you copyable text plus .srt captions.
Drop an audio file, or
Higher accuracy gets names and technical words right more often. The first time it downloads about 240 MB (560 MB with a graphics chip), and it is slower on computers without a graphics chip.
mp3, m4a, wav, ogg, or webm, up to 30 minutes. Runs entirely in your browser, nothing is uploaded. The language is detected automatically; pick it yourself if the result comes out in the wrong language.
Your transcript appears here, ready to copy or download.
How pricing works
Your first 3 transcriptions are free. After that, a transcription costs half a credit from your normal StencilCut credits, so 1 credit = 2 transcriptions and your credits go twice as far as they do on an AI design. Purchased credits never expire; monthly subscription credits refresh each month.
Your audio never leaves your device
Most transcription tools upload your audio to their servers. This one does the opposite: the speech model downloads to your browser once, your device does the work, and the transcript never existed anywhere but your machine. That is why it can be private and free to start.
How it works
Drop an mp3, m4a, wav, ogg, or webm file, or record from your mic. It stays on your device.
Whisper downloads once, is cached, detects the language, and transcribes on your device.
Copy the text, save a .txt, get .srt captions, or send it to Text-to-SVG to engrave.
Frequently asked questions
Is my audio uploaded anywhere?
No. The speech-recognition AI (Whisper) runs inside your browser on your own device. Your audio never touches a server, which is why it is private and costs nothing to run. Only the speech model itself is downloaded, from a public model host, the first time.
Why does the first transcription take longer?
On first use your browser downloads the speech model and caches it. The default Higher accuracy model is about 240 MB when it runs on your processor, or about 560 MB for the faster version that runs on your graphics chip. The Fastest setting downloads about 75 MB (about 200 MB on a graphics chip). Later transcriptions skip the download because the model is already on your device.
Higher accuracy or Fastest: which should I pick?
Higher accuracy is the default and the right choice for lectures, interviews, names and technical words. Fastest uses a smaller speech model: it downloads less and finishes sooner, but it mishears more words, so read the transcript before using it. Both run only in your browser and cost the same. Higher accuracy is several times slower on computers without a graphics chip.
What languages does it support?
The model is multilingual and detects the spoken language automatically, then transcribes in that language. It handles dozens of languages. If a recording comes out in the wrong language (very short or noisy clips can confuse detection), choose the language yourself and run it again.
What files can I use, and can I get captions?
Upload mp3, m4a, wav, ogg, or webm, or record straight from your microphone. You get the plain text to copy or download as a .txt, plus a timestamped .srt caption file for videos.
How much does it cost?
Your first 3 transcriptions are free. After that a transcription uses half a credit, so 1 credit equals 2 transcriptions and your credits go twice as far as they do on an AI design. Purchased credits never expire; subscription credits refresh each month.
Can I turn the transcript into an engraving?
Yes. One button sends the text to the free Text-to-SVG tool, where you can turn a word or phrase into laser-ready vector lettering to cut or engrave.