Loorq

Loorq is a dictation app that also tells you how you sound. Press a key, speak, release. Your words land at the cursor in whatever app you are already in: an email, a doc, a code comment, a Slack message. That part works like good dictation should, on-device and fast enough that you stop thinking about it. The difference is what happens underneath. While Loorq is transcribing, it is also listening for the way you speak. Every "um" and "uh". Every "like" that starts a sentence over. The "you know" you drop in when you are buying a second to think. It counts them, timestamps them, and shows you the shape of how you talked, without ever putting a single one of them into the text you just dictated. WHAT IT ACTUALLY HEARS Filler sounds are the easy part, and Loorq catches them with a purpose-built acoustic model rather than string matching, so a mumbled "uh" is caught even when a transcript would have swallowed it. Crutch phrases are the hard part, because they are real words. Loorq reads context: the "like" that begins a restart, a fluent "you know", an "I don't know" used as a stall, plus "so", "okay", "well", "actually", "basically" and "right" when they are doing filler work instead of real work. WHAT IT REFUSES TO DO Most tools that count filler words are annoying because they are wrong. They flag "I like what I see" and "so much better" and "you know what?" and you stop believing the number. Loorq is built around one rule that never bends: never accuse you of a filler that was a real word. When a phrase is genuinely ambiguous, it undercounts on purpose. The number you see is a floor, and you can trust it. The other rule: your dictated text is yours. Crutch words are shown to you as feedback, never quietly deleted from what you actually said. Loorq cleans up the mechanics of dictation, not your meaning. NOTHING LEAVES YOUR MAC There is no cloud step. The models run locally on Apple Silicon. Your audio lives in memory for as long as it takes to transcribe and is never written to disk unencrypted, never uploaded, never used to train anything. Loorq only ever listens to your own microphone, and only after you ask it to, with a keypress or an explicitly started session. No account required to try it. THE COACHING PART After each take, a small card shows what happened: your words, your pace, your pauses, and any crutches, laid out on the waveform exactly where you said them. Tap one to hear yourself say it. It is a surprisingly effective thing to do once. The dashboard turns that into a trend. Not a vanity score, just whether the thing you are working on is moving. Most people find they have one or two signature crutches rather than a general problem, and those are the ones worth knowing about. You can put any word on a watchlist and Loorq will keep count. People use it for a swear jar, for a hedge they are trying to drop ("I think maybe"), or for a filler that is theirs alone and on nobody else's list. MEETINGS Start a session before a call and Loorq listens the same way, on-device, and gives you the counts afterward. It stores numbers, not words: your meeting is not transcribed and not stored. Practice mode, called Sidecar, is the same loop without the meeting, for a talk you are rehearsing or an interview answer you keep fumbling. IT LIVES IN THE MENU BAR Loorq stays out of the way. Dictation shows up as a small pill near the notch while you speak and disappears when you are done. There is no window to manage and nothing to keep open. TRY IT Three days, every feature on, no account and no card. If you keep it, it is a one-time purchase. No subscription, and updates arrive in the app. REQUIREMENTS macOS 15 or later on Apple Silicon. Loorq needs microphone access to hear you and accessibility access to type at your cursor, and it asks for each one at the moment it first needs it.