Subformer
Free · No signup · Runs in your browser

Free Speech to Text in 35+ Languages

Drop any audio or video file, pick a language, and get a full transcript with timestamps - all in your browser. The model is downloaded once (~75 MB) and cached locally, so subsequent runs are instant and work offline.

Private by default

Recognition runs locally. Your audio never leaves your device.

Instant after first load

The model is cached in your browser. Subsequent runs work offline.

Free, no signup

No account, no credits, no usage cap. Open and use.

Choose a language

Under the hood

Speech recognition normally means uploading your audio to someone's GPU. Here the model runs on your own machine instead, which is what makes the tool free and private at the same time.

Open-source recognition model
A quantized build of an open-source neural speech recognition model, small enough to download once and keep in your browser.
WebAssembly
The inference runtime is compiled to WebAssembly, so the numerical code runs at near-native speed inside a sandboxed tab with no plugin and no install.
On-device inference
Loading the model, tokenizing, and running the decoding loop all happen client-side. No part of the pipeline calls out to a server once the page has loaded.
Web Audio API
Your file is decoded and resampled to the 16 kHz mono the model expects using the browser's own audio pipeline, so any format your browser can play is accepted.

The trade-off is model size. On-device weights are a fraction of what a hosted service runs, so accuracy drops on noisy audio, strong accents, and overlapping speakers - that is where the cloud option below earns its keep.

Need higher accuracy or more languages?

Cloud-based Speech to Text

The free in-browser tool runs a small open-source model - fine for clean audio, but it can struggle with heavy accents, noise, or overlapping speakers, and only covers a limited set of languages. For broadcast-grade accuracy, far broader language support, and SRT/VTT/TXT downloads, use our paid cloud-based Speech to Text.

Try cloud-based Speech to Text