AudioKit

Transcribe audio to text

Get the words out of an interview, lecture, voice memo or video. Then have AI turn the transcript into a summary, show notes or clean prose.

Free, up to 10 minNo sign-upSRT subtitlesAsks before sending audio

Drop a voice memo, interview, lecture or video

Up to 10 minutes per transcription. You can also drag a file here or paste it.

How it works

  1. On your device: the audio track is extracted and squeezed to 32 kbps mono at 16 kHz. Speech models only use frequencies up to about 8 kHz, so nothing useful is lost, and a 10-minute file drops to about 2.4 MB.
  2. Speech recognition: that small file goes to Google’s Gemini transcription model through Vercel AI Gateway, which returns the text and, when available, per-segment timing.
  3. Optional writing help: “Turn it into…” sends only the text to an OpenAI language model with strict instructions to use nothing but the transcript.

Get a better transcript

  • Record close to the mic; room echo hurts accuracy more than background hum.
  • For long recordings, remove silences first so more speech fits into 10 minutes, or split it with the trimmer.
  • Very quiet? Normalise it before transcribing.

Questions people ask

How do I transcribe audio to text for free?

Drop the recording above, tick the consent box and press Transcribe. Up to 10 minutes per file is free with no account. You get editable text, optional timestamps, and .txt or .srt downloads.

Is my audio uploaded?

For this one tool, yes, a copy is sent, because speech recognition models are too large to run in a browser tab. Before sending, AudioKit compresses the audio on your device to mono 16 kHz speech quality, and it only sends after you tick the consent box. The audio and transcript are not stored. Every other AudioKit tool runs fully offline.

Which languages are supported?

The model detects the language automatically and handles English and dozens of other major languages, including Spanish, French, German, Portuguese, Hindi, Japanese and Chinese. Accuracy is best on clear speech with little background noise.

Can I get subtitles?

Yes. When the model returns timing, you can download an .srt subtitle file that YouTube, Premiere, DaVinci Resolve and VLC accept.

What does “Turn it into…” do?

It sends the transcript text (not the audio) to a language model that writes a summary, show notes with timestamps, a cleaned-up version without filler words, or a list of quotable lines. It only uses what is in the transcript.

How accurate is it?

Modern speech models get most words right on clean recordings. Names, jargon, heavy accents and crosstalk cause most mistakes, so read the transcript before you publish it. You can edit the text directly in the box.

More tools on the bench