Audio to Text — Transcribe a Recording, Online Free
Turn recordings into text in 90+ languages, with timestamps — right in your browser.
Drop an audio or video file, or click to browse
MP4, MOV, WebM, MP3, WAV, and more · up to 10 minutes · first run downloads the model, then about as long as the audio
Processed locally — your file never leaves this tab
The first run downloads the model once (Tiny ≈ 40 MB, Base ≈ 80 MB) and your browser caches it — later runs work offline.
Processed locally — your file never leaves this tab
Drop in a recording and get the words out of it: voice memos, interviews, lectures, podcasts — in about 90 languages, detected automatically when you are not sure which one it is. A Whisper AI model runs inside your browser tab on your own device, so the audio is never uploaded, and the transcript comes out as editable text or an SRT file with timestamps.
Transcripts people actually need
- Turn a lecture, meeting or interview into notes you can search and quote.
- Caption a video or podcast episode by exporting an SRT with timestamps.
- Pull the words out of a voice memo without sending the recording anywhere.
How to use the Audio to Text
-
Drop your file
Add an MP3, WAV, M4A, video file or other recording. It is decoded on your device and never uploaded.
-
Pick a language and model
Auto-detect handles most files. Choose Tiny for speed on long recordings or low-end devices, Base for the best accuracy.
-
Transcribe and download
The first run downloads the model once and caches it. Edit the transcript if you like, then save it as TXT or SRT.
A local transcript, not a live caption
Audio to text runs a Whisper model in the browser. The model downloads once from a CDN, then the file is transcribed on the device. The audio is not uploaded for the transcript.
The cap is about 10 minutes. There are no speaker labels. It does not listen to the microphone live. Record first on the voice recorder, export, then drop that file here.
Model and language
About 90 languages. Pick the one you spoke, or leave detection on if you are not sure:
- Tiny is the smaller download and enough for clear speech.
- Base is slower and a bit more accurate on noisy or accented speech.
- WebGPU is faster on machines that have it. Otherwise it falls back to the CPU.
What the text will get wrong
Names, overlapping talk, and music under a voice come out messy. It will not mark who spoke. Clean the file first if the ends are long silence: trim, or run the silence remover, then transcribe the shorter file.
You can edit the words before download, as plain text or as an SRT with timestamps. It does not translate them into another language.
Audio to Text FAQ
Is my audio uploaded?
No. The file is decoded in your browser and transcribed by a Whisper model that runs locally through WebAssembly or WebGPU. The only network request is the one-time model download from a public CDN — after that, transcription itself works offline. Nothing about your audio ever reaches a server.
How accurate is the transcription?
It uses OpenAI’s Whisper models, which score near human level on clear speech. Accuracy drops with heavy background noise, strong accents, overlapping speakers or loud music. The Base model is noticeably better than Tiny on difficult audio — and either way the transcript lands in an editable box, so you can fix names and punctuation before downloading.
What are the limits?
One file at a time, up to 10 minutes per pass — trim longer recordings first (the audio trimmer works the same private way). It does not label speakers: a dialogue comes out as one flowing text without names. And this page works on files, not the live microphone.
How is this different from the voice recorder?
The voice recorder captures your microphone into an audio file; this page reads an existing file and writes out the words. They pair well: record there, transcribe here. Both run entirely in your browser.
Which languages does it support?
Whisper was trained on about 90 languages, from English, Chinese, Spanish and Arabic to Vietnamese, Ukrainian and Swahili. Leave the language on Auto-detect and the model works it out from the audio, or pin a language for a small accuracy boost.
Does it work on my phone?
Yes — recent Chrome, Edge, Firefox and Safari, desktop or mobile. Browsers with WebGPU (Chrome, Edge) transcribe much faster; elsewhere it runs on the CPU, so pick the Tiny model for long files and expect a wait of a few minutes on ten minutes of audio. The one-time model download is smaller for Tiny.
Related tools
Voice Recorder
Record your microphone, trim the take, save it as MP3 — nothing leaves your device.
Open Voice Recorder
Audio Extractor
Save the soundtrack of any video as MP3 or WAV — locally, in seconds.
Audio Trimmer
Trim audio online — free, right in your browser. Nothing is uploaded.