Speech to text (Whisper in the browser)
Transcribe audio or video with OpenAI's Whisper running in your tab, as plain text or SRT subtitles.
Or choose them. Files are processed in this tab and never uploaded to any server.
Whisper, released by OpenAI as openai/whisper, is the open speech model most transcription tools are built on. This page runs it in the tab with huggingface/transformers.js: the browser decodes the audio track of your recording, resamples it to 16 kHz mono, and the model transcribes it in 30-second windows. Choose tiny for speed or base for accuracy, set the language if you know it, and get plain text or an SRT file with timestamps. It is free and private, and it is slower than a cloud API.
How it works
- The quantized models are downloaded from Hugging Face on first use, about 40 MB for tiny and 77 MB for base, and cached by the browser.
- Long files are split into 30-second chunks with 5 seconds of overlap, the standard way Whisper handles audio longer than its window.
- Translate mode makes Whisper write English whatever language is spoken; Whisper has no other target language.
Where your data goes
Nowhere. This tool runs entirely in your browser: the text you paste is processed by the page and is never transmitted to a server or written to a log.
This tool is free and needs no account. Its results exist only in your open page and are not saved anywhere.
What it costs
This tool is free, with no sign-in and no points.
Common questions
- How long will it take?
- It depends on your CPU, because the model runs on WebAssembly rather than a data-centre GPU. Expect the base model to take a sizeable fraction of the recording's length on a recent laptop, and much longer on a phone. For hours of audio, a cloud API or whisper.cpp on your own machine is the practical choice; this page suits clips and meetings.
- The model download never finishes. Why?
- The weights come from huggingface.co and the ONNX Runtime files from jsDelivr. Both can be slow or unreachable on some networks, including in mainland China, and the tool cannot start until the download completes. Once it has worked once, the browser cache lets it start without the network.
- Which files can it read?
- Anything your browser can decode: MP3, WAV, M4A, AAC, OGG and FLAC, plus the audio track of MP4 and WebM video in most browsers. If decoding fails, extract the audio first — for example with ffmpeg -i input.mov -vn audio.m4a. Files are capped at 100 MB because the whole decoded track is held in memory.
The open-source behind it
This tool runs on huggingface/transformers.js, released under Apache-2.0. If you need the same behaviour inside your own program, that is the library to reach for.
huggingface/transformers.jsAlso known as
- speech to text
- audio to text
- whisper online
- transcribe audio free
- video to srt
- whisper browser