Runs on your machine · nothing uploaded
Transcribe the recordings you are not allowed to upload.
TapeText runs a speech model inside your browser. Drop in an interview, a lecture, a deposition or a call recording, correct it against the audio, and export subtitles or a timecoded document. The file never leaves your computer, which is the point.
checking your browser
No upload, and you can check
The model is fetched to your browser and runs there. Open the Network tab while it works, or transcribe a second file with your wifi off: once the model is cached, the tool does not need a connection at all.
No cap, because there is no bill
Every service that transcribes on its own servers pays per minute, and every one of them caps the free tier because of it. Nothing is metered here: an afternoon of interviews costs the same as a voice memo.
An editor, not a text dump
Every word carries its own timestamp. Click one to jump the audio there, fix what was misheard, split a paragraph where the speaker actually paused, and label who is talking.
What it is honestly not
Four things, stated here rather than discovered on your own recording.
It does not separate speakers for you
You apply labels by hand, and a label runs forward until you change it. Automatic separation is unreliable on crosstalk and phone audio, and a confidently wrong attribution in a transcript is worse than a blank one.
It does not detect the spoken language
You pick it before you start. The library underneath does not detect language; it silently assumes English, so a picker claiming otherwise would turn a French interview into fluent nonsense.
It is not usable on a phone
The model wants about a gigabyte of memory and minutes of uninterrupted work. The page checks your browser and says so before anything downloads.
It does not burn subtitles into video
It writes SRT and VTT files you add in your editor. Rendering text into a picture is a video encoding job, and a different tool.
Written for a particular problem
Transcribe audio without uploading it
Transcribe interviews and recordings with nothing leaving your computer. The speech model runs in your browser. Free, no account, no file size limit.
Transcribe an interview without handing it to a service
Transcribe recorded interviews with speaker labels and timecodes, without sending the audio to a service. Built for journalists and researchers. Free.
Make an SRT from a video, in the browser
Turn MP4, MOV or WebM into an SRT or VTT subtitle file in your browser. No upload, no watermark, no minute cap. Edit the text before you export.
Offline transcription, with nothing to install
Transcribe audio to text with no internet connection. Load the page once, then work offline on a plane or an air-gapped laptop. Free, nothing to install.
Free transcription with no account and no minute allowance
Transcribe audio to text free, with no account, no credit card and no monthly minute allowance. Runs in your browser, so there is no per-minute cost to cap.
Questions
- Is TapeText free?
- Yes: no account, no sign-up and no limit on how long or how many files you transcribe. A one-time paid add-on is planned later for batch processing, a larger model and formatted document export; everything the tool does today stays free.
- Is my recording uploaded anywhere?
- No. The speech model is downloaded to your browser and runs on your own machine. You can verify it in the Network tab of your developer tools, or by transcribing a second file with your connection switched off.
- How accurate is it?
- It is OpenAI's Whisper, the same model most transcription services are built on, in its small and base sizes. On clean speech it is good enough to read; on accents, crosstalk and phone audio it makes mistakes, and names and numbers are where it fails most. Everything is editable, and clicking a word plays it.
- How long does it take?
- Several times faster than real time on a machine with a graphics adapter, roughly five to twelve minutes for an hour of audio depending on the model. Without one it runs on the processor and is several times slower again; the page tells you which it is using before you start.
- Does it work on a phone?
- Not well, and the page says so rather than letting you find out. Recognition needs more memory and more sustained compute than a phone browser will give it. A laptop or desktop with Chrome, Edge or another Chromium browser is what this needs.
- Does it separate speakers automatically?
- No. You label speakers yourself, and a label carries forward until you change it. Automatic separation is unreliable on exactly the recordings that matter (crosstalk, phone lines, a single room microphone), and a confidently wrong attribution is worse than none.
- Which languages does it handle?
- Ninety-nine. You pick the spoken language before you start; there is no automatic detection, and the picker says so rather than implying one. It can also translate what it hears into English instead of transcribing it in the original language.
- What can I export?
- SRT and VTT subtitle files, plain text in paragraphs, and Markdown with a timecode on every paragraph for anything you need to cite.