No upload, 100% local, no account

Audio Transcribe, speech to text in your browser

Drop a recording and get a text transcript. Runs on your device with a small Whisper model downloaded once from this site; your audio is never uploaded.

How Audio Transcribe · speech to text works

Audio Transcribe turns a recording into text using Whisper, a speech recognition model that runs entirely inside your browser tab through Transformers.js. The first time you run it, your browser downloads the model and the WebAssembly engine it needs (about 68 MB total) from this site, then caches them; every transcription after that happens on your device with no server involved and no audio ever leaving your machine. That is the same trade-off as the OCR and audio tools on this site: a real one-time download for real offline processing, not a marketing claim.

The model used here is whisper-tiny.en: the smallest, quantized, English-only Whisper variant. It is fast enough to run in a browser tab and produces a useful first draft, but it is not a professional transcription service. Expect mistakes on proper nouns, technical vocabulary, strong accents, background noise and overlapping speakers. Recordings longer than about 30 seconds are processed in overlapping chunks so the whole file gets transcribed, but a one-hour interview still takes real computing time since it all runs on your own device rather than a data center.

Written by Bastien Sulyan

How to use Audio Transcribe · speech to text, step by step

  1. Drop an audio file (MP3, WAV, M4A, OGG or the WebM recordings this site's Voice Recorder produces).
  2. Wait for the one-time engine and model download (about 68 MB) if this is your first transcription on this device.
  3. Choose plain text (.txt) or timestamped subtitles (.srt) as the output format.
  4. Press transcribe and let Whisper process the recording fully in your browser tab.
  5. Copy the text inline or download the .txt/.srt file once it is ready.

Common use cases

  • Getting a rough written draft of a voice memo recorded on your phone before cleaning it up by hand.
  • Turning a short recorded interview into searchable text without uploading it to a cloud transcription service.
  • Generating a starting-point .srt subtitle file for a short clip, to be corrected afterward.
  • Recording a quick voice note with this site's Voice Recorder, then transcribing it in the same browser tab.
  • Getting a first-pass transcript of a lecture or meeting recording when perfect accuracy is not required.

Frequently asked questions

How accurate is the transcription?

It depends heavily on audio quality. Clear, single-speaker English recorded close to the microphone with little background noise transcribes reasonably well. Accents, cross-talk, background music or noise, and specialized vocabulary (names, jargon, acronyms) all increase errors. This is a small, quantized model chosen so it can run inside a browser tab, not the largest Whisper model, so treat the output as a useful draft to review and correct, not a final transcript.

Does it work with languages other than English?

No. This tool uses whisper-tiny.en, the English-only Whisper variant, chosen because it is smaller and more accurate on English than the multilingual tiny model. Audio in other languages will produce poor or nonsensical results. Support for other languages is not currently available.

Why does it need to download something before it works?

The speech recognition model and the WebAssembly engine that runs it (about 68 MB combined) are not built into the page; they are fetched once from this site and cached by your browser, similar to how the OCR tool downloads a language model. After that first download, transcription runs fully offline. Nothing is fetched from any third-party CDN.

Is my audio uploaded anywhere?

No. The recording is decoded and processed entirely inside your browser tab using the Web Audio API and a local Whisper model. No audio file, and no transcribed text, is ever sent to a server. You can disconnect from the network after the model has downloaded and transcription still works.

How long does transcription take?

It depends on your device's CPU and the recording length, since everything runs locally rather than on server hardware. Short clips (under a minute) typically finish in a few seconds to under a minute. Long recordings, such as a one-hour interview, take proportionally longer and will keep your browser tab busy for that duration; keep the tab open and do not navigate away while it works.

What is the difference between the .txt and .srt output?

The .txt file is the plain transcribed text with no timestamps, useful for reading or pasting elsewhere. The .srt file is a standard subtitle format with a timestamp range for each spoken segment, useful for adding captions to a video. Both are generated from the same transcription; pick whichever fits what you need to do next.