RunTangent
AUDIO & VIDEO / AUDIO TRANSCRIBER

Audio Transcriber

Transcribe audio into text with a model running in your browser.

Keep the words. Stay with the thought.

Your listening deskEnglish recognition · Editable text · Local processing
THE RECORDINGBring something worth listening to.Audio up to 30 minutes / 100 MB, subject to browser decoding limits.
Load audio
THE WORDSMake room for what was said.

A place for the words.

Load a recording to begin.
Your transcript will find its home here.

English speech recognition. Review the result with care.

Speech recognition runs in this browser. Model and runtime downloads require a network connection. Results can contain mistakes.

FROM SPOKEN TO WRITTEN

Listen once.
Return to the words.

Make interviews, voice notes and spoken ideas easier to revisit. Keep the source nearby, review the recognition, and shape a transcript that says what you meant.

How to transcribe a recording

  1. Bring your recording.Choose or drop a supported audio file. Short, clear clips are a good starting point.
  2. Prepare the engine.Review the model download notice and choose Load model & transcribe.
  3. Listen and refine.Play the source and check the completed text. Recognition is a starting point, not a perfect record.
  4. Keep the words.Copy your edited transcript or download a plain-text file before clearing or leaving.
A FEW WORDS BEFOREHAND

Room for clarity.

Back to the workspace ↑
Does my audio leave this browser?

This tool decodes and transcribes your file in the browser; it does not upload the recording. After your consent, it loads a speech library, runtime and model files from external hosts. Browser caching may reduce downloads but is not guaranteed, so an internet connection may still be needed.

Which recordings can I use?

Choose an audio file your browser can decode, such as supported MP3, WAV or M4A recordings. Codec support varies. Files must be no larger than 100 MB and no longer than 30 minutes; the decoded sample limit can require shorter clips.

Which language is supported?

The current model is Whisper Tiny English. Use clear English speech. Recognition can make mistakes, especially with noise, accents, names and overlapping speakers. Review every transcript before relying on it.

Why do I need to load a model?

Recognition runs on your device using model files that may total roughly 70 MB, plus runtime assets. Loading and processing time depend on your network, browser and hardware. The tool asks for consent before requesting the speech engine.

Can I edit and save the transcript?

Yes. Review and edit the completed transcript, then copy it or download a plain-text file. There are no speaker labels, timestamps or subtitle exports in this version. Clear / Cancel stops applying pending results; model loading or inference already underway may continue until it settles.