Speech to Text

Speech to Text offers two independent engines for the same goal. Real-time mode taps the Web Speech API — the SpeechRecognition interface built into modern browsers — and streams words onto the page as you talk: interim guesses first, then a finalized line once you pause. File upload mode instead loads a Whisper AI model through Transformers.js and transcribes a recording you already have, without listening to a microphone at all.

Reach for real-time dictation when you want to capture thoughts hands-free: meeting notes, a rough draft, or a quick memo you would rather speak than type. It shines during short, live sessions where seeing text appear immediately keeps you in flow.

File upload is the better fit for material that already exists — an interview, a lecture, a voice memo, or the audio track of a video. Once the Whisper model finishes its one-time download, that mode keeps working even with no connection, which suits long recordings and repeated jobs.

For live work, pick your spoken language from the dropdown, press Start Recording, and grant microphone access when the browser asks; your speech fills the transcript and you can edit it afterward. For a recording, switch to File Upload, choose a model — Tiny for speed, Small for stronger multilingual accuracy — and drop in an MP3, WAV, MP4, or similar file. After the model runs, the finished text is ready to copy or download.

Real-time recognition depends on the browser: Chrome, Edge, and Safari implement the Web Speech API, while Firefox does not, so use File Upload there. Recent Chrome versions can install an on-device language model, keeping microphone audio local; otherwise real-time speech is handled by the browser vendor's speech service. The real-time dropdown lists 22 languages, whereas Whisper covers 99+ with automatic detection. File uploads accept common audio and video formats up to 500MB; for video, pulling the audio out first speeds things up and improves the result.