Speech to Text

Typing slows you down when ideas are flowing, and some thoughts are just easier to say than to type. This Speech to Text tool turns your spoken words into written text in real time, so you can capture a note, draft an email, or dictate a paragraph hands-free.

It uses your browser's built-in speech recognition. Press the microphone button and the transcript fills in as you talk, with words appearing live before they are finalized. You can pick the recognition language — US or UK English, Spanish, French, German, Hindi, and more — so it understands your accent and vocabulary. Journalers, people with repetitive-strain concerns, and anyone brainstorming out loud use it to get words down fast, then copy the transcript or clear it and start again.

It is worth knowing how this works under the hood: browser speech recognition typically sends your audio to an online service for transcription, so this is best for everyday notes rather than confidential material.

This uses the browser's Web Speech API for recognition, and it is important to be clear about where the work happens. In Chrome — and in most browsers that implement it — the audio captured from your microphone is sent to the vendor's speech recognition service and transcribed there, then the text is returned. It is not processed locally.

That makes this one of the small number of tools on this site where your data leaves your machine, and it is listed as such on the methodology page. Do not use it to transcribe anything confidential: a private conversation, medical or legal material, or anything covered by a confidentiality obligation.

Support is also uneven. The API is unevenly implemented across browsers and may be absent entirely, in which case the tool cannot function. Microphone access requires your explicit permission, and a secure connection.

Recognition quality depends more on the recording than on the engine. A close, consistent microphone position in a quiet room outperforms any amount of post-editing, and hard surfaces that produce echo are the most common cause of poor results at home. Speaking at a normal pace and articulating clearly helps considerably; speaking unnaturally slowly does not.

Expect to correct proper nouns, technical vocabulary, and acronyms, which are where general-purpose recognition is weakest because they are rare in its training data. Punctuation is inferred from pauses and intonation and is frequently wrong, so a transcript usually needs a punctuation pass regardless of how clean the words are. Accents and dialects are handled unevenly, and selecting the closest available language variant is worth doing.

Two notes on limitations. Multiple speakers are not distinguished — separating who said what is speaker diarisation, a separate capability this does not offer — so an interview arrives as one undifferentiated block. And for anything long or important, a dedicated transcription service with an editing interface will be faster overall than correcting a raw stream by hand.