Hands-Free Writing and Dictation
Speech to Text (STT) converts spoken language into digital text. Students can use this tool to dictate lecture drafts, record notes hands-free, or quickly draft long essays. By bypassing the keyboard, dictation allows ideas to flow naturally and reduces typing strain.
SpeechRecognition under the hood:
The tool utilizes the Web Speech Recognition API (SpeechRecognition and webkitSpeechRecognition). When you tap the microphone, the browser requests permission to access your audio hardware, transcribes the incoming soundwaves into text packets, and renders them in the editor.
The transcription text remains editable, allowing you to manually add punctuation, adjust words, or format layouts before copying the final draft.
When dictation is faster than typing, and when it isn't
Dictation wins on first drafts and loses on edits. Speaking runs at roughly 130 words a minute against 40 typed, so a lecture summary or an essay's opening argument comes out several times faster — but only while you are producing new prose in long runs. The moment the work turns into fixing a clause in the third paragraph, the keyboard is faster again, because navigating to a position by voice costs more than the correction saves. The practical pattern is to dictate a whole section without stopping to fix anything, then switch to the keyboard and clean it up.
The other consideration is what you are writing. Continuous prose transcribes well. Anything with dense notation — code, equations, chemical formulae, tables of figures — does not, because the recogniser is a language model of everyday speech and has no way to know that "x squared" should come out as a symbol rather than three words.
What determines accuracy
Four things, in roughly this order of impact. Background noise matters more than microphone quality; a laptop mic in a quiet room beats a good headset in a cafe. Speaking pace — recognition works on phrases rather than single words, so a steady, even rhythm gives the engine more context than either rushing or over-enunciating word by word. The language setting must match what you are actually speaking; a mismatch degrades results far more than an accent does. And vocabulary: proper nouns, technical terms and acronyms are the most common errors, because they are rare in the model's training data.
Punctuation is not inferred. Say "comma", "full stop" and "new paragraph" explicitly, or add them afterwards in the editor — the transcript stays fully editable for exactly this reason.
Browser support
This uses the browser's own Web Speech Recognition API, so availability is the browser's decision rather than the tool's. Chrome, Edge and other Chromium-based browsers support it; Safari's support is partial and Firefox does not implement it at all. Where it does run, the recognition itself may happen on the browser vendor's servers rather than on your device — that is the browser's implementation detail, not something this page controls, and it is worth knowing before dictating anything confidential.