Audio to text

An interview, a lecture recording, a voice memo capturing an idea on the go: this tool listens to the recording and writes it down as searchable, copyable text. The recognition model is OpenAI's Whisper, here as a purpose-compiled, single-thread WebAssembly build - the same underlying technology well-known transcription services use, except it runs entirely on your device instead of a foreign server.

By default, Whisper detects the spoken language itself ("Auto-detect") - for short or multilingual recordings, the language can also be fixed explicitly, which can slightly improve accuracy and speed. The result appears as running text with a paragraph per speech segment; for videos that need embedded timestamps, the sibling tool "Video to subtitles" uses the same recognition but outputs an SRT/VTT file with exact time marks instead.

Because there is no shared memory (no cross-origin isolation), recognition runs single-threaded rather than multi-threaded - correct, but noticeably slower than a server-side service. That is why a much stricter length cap applies here than for the site's other audio tools; honestly slow beats promising a false speed.

Specifications

In 3 steps

1. Drop your audio file.
2. Pick a language or leave it on "Auto-detect".
3. Copy the recognised text or download it as TXT.

Limitations: Runs single-threaded (no cross-origin-isolation mode on this site) and is therefore noticeably slower than server-side services - a much stricter per-device length cap applies here (shorter than the other audio tools). Uses the compact "tiny" model tier: good for clear speech, weaker with heavy background noise, dialects or jargon than a larger model would be. Recognition is not error-free - check the result, especially proper names and numbers.

FAQ

Related tools

Video to subtitles · Remove noise · Image to text (OCR) · Create a waveform image · Record audio · Swap left/right