Lunosia — private transcription

Lunosia turns speech into text without any of it leaving your computer. There is no account, no upload and no server. The speech recognition model runs inside the extension, on your own machine. Once it has downloaded you can disconnect from the internet entirely and it still works — which is the simplest proof that nothing is being sent anywhere. WHAT IT DOES • Transcribe your microphone, live, as you speak • Transcribe audio from any tab — calls, lectures, videos, podcasts • Transcribe recordings you already have — mp3, m4a, wav, mp4 • Speaker labels, so interviews and meetings read as conversations • A summary tab that pulls out key points, dates, and what you owe • Export as .txt, .md, .srt or .vtt — subtitles ready for any editor • 99 languages, and it can write any of them out in English BUILT FOR ROOMS THAT ARE NOT QUIET There is a setting for when the speaker is across the room. Browsers normally strip out distant audio as noise, which is exactly wrong for a lecture hall or a meeting room, so Lunosia turns that off when you ask it to. It also handles audio with music underneath it, which most transcription tools quietly drop. WHY THIS RATHER THAN A CLOUD SERVICE Every other transcription tool uploads your audio. For a lecture that may not matter. For a medical appointment, a legal call, a therapy session, an interview with a source, or anything under NDA, it matters a great deal — and "we delete it after 30 days" is a promise you cannot verify. Lunosia makes the question moot. There is no server behind this extension. Not a private one, not an encrypted one — none. Your transcript is in your browser and nowhere else. The research links in the summary tab work the same way: they are built on your machine from words already in your transcript. Clicking one is an ordinary search, so the single word you click is all that ever leaves the computer. PRICING The free version transcribes your microphone, in every language, with .txt and .md export. Lunosia Pro is one payment of $19 — no subscription, no account, no server. It adds tab capture, transcribing existing recordings, the summary tab, speaker labels, .srt and .vtt subtitles, and the more accurate model for names and technical words. PERMISSIONS Lunosia asks for access to huggingface.co and nothing else. That is where the speech model is downloaded from, once. It requests no access to your browsing, your tabs, your history or any website.