VoiceDumps - Hold the globe key. Talk. The words appear at your cursor.

by
Local speech to text for macOS. Hold the globe key, talk, and the words land at your cursor in whatever app you were already in. Or drop in audio and video for transcripts with real timestamps. Free, MIT, and fully offline — the models ship inside the app.

Add a comment

Replies

Best
Hey Product Hunters 👋 I built VoiceDumps because I kept hitting the word cap on cloud dictation, and because I did not love that every voice note I made was a round trip to somebody's GPU. Hold the globe key, talk, let go — the text lands at your cursor in whatever app you were already in. No account, no API key, nothing uploaded. The speech models ship inside the .app, so a fresh install works on a plane, on first launch. The thing I most want picked apart is the latency claim. On an M1 Pro, the worst of three runs from "stopped speaking" to "text on the clipboard" was 429 ms on a ~6-second utterance. The best-known cloud dictation app publishes "under 700 ms" as a p99. So: their p99 against my worst case, on the smaller of my two models. It stops being true past about ten seconds of speech, where they win — and their budget buys an LLM pass that reformats your text, which I do not do. I hand you what you said. Rather than ask you to believe any of that, scripts/bench.sh is in the repo. Run it on your machine and tell me what you get. Two things I would genuinely like opinions on: 1. It is a 720 MB download, because two quantised Whisper models are inside the app instead of an API key. Is that a fair trade for working offline, or a dealbreaker? 2. Apple Silicon only right now. If you are on Intel or Windows, is that a "shame" or a "come back when"? Free, MIT, no telemetry, no upsell — there is no paid tier to graduate you onto. github.com/heynaavi/voiceDump One setup gotcha, because three people will otherwise find it independently: set System Settings → Keyboard → "Press 🌐 to:" → Do Nothing, or macOS opens the emoji picker underneath the app.