Skip to content
TabShed
My storage

Audio to Text Converter – Free Transcription

Turn a recording, podcast, lecture or video into text with AI speech recognition. Free, no sign-up, and the file never leaves your device.

Processed on your device · no upload

Drop an audio or video file to turn it into text

MP3, WAV, M4A, MP4, MOV, MKV, WebM and more · up to 500 MB · or paste with Ctrl+V

Runs on your device · nothing is uploaded. The first run downloads a speech model once (41–249 MB, depending on the quality you choose).

What to do next

Subtitle GeneratorMake timed SRT and VTT subtitles from the same fileVideo to AudioPull the audio out of a video firstAudio CutterCut a long recording to just the part you needAudio NormalizerEven out a quiet recording before you transcribe it

How to use Audio to Text

  1. Add your audio or video

    Drop an MP3, WAV, M4A, MP4, MOV or WebM file (up to 500 MB). Videos are fine: only the sound is read.

  2. Pick quality and language

    Fast is best for clear English, Accurate for noisy audio or other languages. Leave the language on automatic if you are not sure.

  3. Press Transcribe

    The first run downloads the speech model once (40–250 MB). You see the text appear as each 30-second part is finished, and you can cancel any time.

  4. Copy or download

    Edit the text if you like, then copy it or save it as TXT or Word. Need timed subtitles instead? Use the Subtitle Generator.

Frequently asked questions

Is my audio uploaded to a server?
No. The speech model runs inside your browser, on your own processor or graphics card. Your file never leaves your device. The model itself is downloaded once and stored in the browser cache, so later runs also work offline.
Which AI model does this use?
OpenAI Whisper (tiny, base and small), MIT licence, in the ONNX versions by the Hugging Face community (onnx-community/whisper-*), run with Transformers.js (Apache-2.0) and ONNX Runtime Web (MIT). Model files come from the Hugging Face Hub.
How accurate is it, and how long does it take?
Clear speech in English is usually very good, even with the Fast model. Accents, noise, music and technical words work better with the Accurate model. On a laptop expect roughly real-time to a few times faster than real-time with a GPU; without a GPU a one-hour file can take a long while, so keep the tab open.
Which languages are supported?
Whisper understands about 99 languages. The list here shows the most common ones; automatic detection works for the rest, but the Accurate model is much better outside English.
What can it not do?
It does not tell speakers apart, it does not add punctuation perfectly, and it can invent words in silence or background music. Always read the result before you use it for something important. Very large files (over 500 MB) and phones with little memory may fail.
Can I get subtitles with times?
Yes: use the Subtitle Generator for SRT and VTT with editable timing. Here you can also download an SRT after the text is ready.

More audio tools

Audio CutterCut and trim audio filesAudio JoinerJoin several audio files into oneVolume BoosterMake MP3 and audio files louderPitch ChangerShift pitch up or down by semitones