Audio to Text
Record straight from your microphone or drop in files you already have — several at once if you like — and get the text back with timestamps you can export as subtitles. Speech recognition runs on your own device, so nothing is uploaded. That matters for interviews, lectures and anything confidential.
Add audio files
or drag and drop them here
Add one or more audio files, or record straight from your microphone.
Try next
Your files are processed locally in your browser. They are not uploaded to our servers.
How to use
- Add your audio files — MP3, WAV, M4A, OGG, FLAC and most other formats work — or hit “Record now” to capture straight from your microphone.
- Pick a model: fast for a quick draft, accurate for the best result. Set the spoken language if you know it, or let it be detected.
- Start the transcription. Each file gets its own transcript to copy or download as TXT, SRT or VTT — or grab everything at once as a ZIP.
What is this tool?
Audio to Text runs OpenAI's Whisper speech recognition model directly in your browser through WebAssembly and WebGPU. The model is downloaded once from a public CDN and cached afterwards; your recording itself is never sent anywhere. That is the whole point: transcription is exactly the kind of task where the file is sensitive — job interviews, therapy sessions, medical notes, legal recordings, unpublished journalism — and every mainstream alternative asks you to upload it to a server first. Because the work happens locally there is also no minute quota, no watermark and no account. Whisper is multilingual and transcribes dozens of languages, and it can translate what it hears into English in the same pass. Files are processed one after another with the model loaded only once, so a batch costs little more than a single recording.
Common uses
- Turning an interview or meeting recording into searchable text
- Producing SRT or VTT subtitles from a video's sound track
- Writing up a lecture or podcast episode without typing it out
- Getting an English transcript of a recording in another language
FAQ
Is my recording uploaded anywhere?
No. The speech model runs inside your browser, so the audio never leaves your device. The only thing downloaded is the model itself (about 40–250 MB depending on which one you pick), which your browser then caches for later runs.
How long does it take?
It depends on your browser. Chrome and Edge on a desktop use WebGPU and manage a fraction of the recording's length. Safari and older browsers fall back to a slower mode where transcription takes roughly as long as the recording itself. The smaller model is considerably faster if you just need a rough draft.
Does it show who is speaking?
No. The transcript is one continuous text; it is not split by speaker. Separating voices needs a second, different model and is not part of this tool — worth knowing before you use it on a multi-person interview.
Can I transcribe a video?
Not directly — this tool takes audio files. Pull the sound track out first with the Video to MP3 Converter, then transcribe that. The timestamps you get back still line up with the original video, so the subtitles fit.