The words showed up eventually — just not in half a second

A gap of a few seconds, or occasionally longer, between letting go of the hotkey and the text actually appearing is the specific complaint here, and it's worth separating from every other kind of "slow": the microphone worked, the pill and level meter appeared normally, and the transcript did eventually land — it just took longer than expected to get there. On an older Mac this is common enough to have a specific set of usual causes, most of which are fixable without replacing anything.

What "normal" actually looks like, so slow is easier to recognize

On typical hardware, a short dictation appears roughly half a second after the key comes up — closer to instant than to a wait. If what's showing up instead is several seconds, or a wait that seems to grow the longer the dictation runs, that's outside normal range and worth diagnosing rather than living with.

That timeline matters here specifically because it means a slow dictation is an annoyance, not a risk: the transcript is already saved before any processing that could be slow even begins.

The one-time cause: the first dictation after installing

The very first time the app is asked to process speech, it fetches its engine over the network — a one-time event, not something that happens on every dictation afterward. If the very first attempt after installing feels slow, that's this fetch happening in the background, and it depends on the network connection at that moment rather than on the Mac's age at all. Every dictation after that first one runs entirely locally, with no network step in the path, so a slowdown that persists past the first use has a different cause than this one.

The cause most often responsible on genuinely older hardware

Coii VoiceInput's processing runs locally, on the Mac itself, rather than being sent anywhere — that's the whole basis of working with the network off. The honest tradeoff of that design is that speed scales with how much the Mac's own processor and memory can do, and an older Mac, particularly an older Intel model near the low end of what macOS 13 Ventura still supports, simply has less of both than a current one. This isn't a bug to fix so much as a fact to plan around: on the oldest supported hardware, a longer wait after release is the expected shape of the tradeoff, not a malfunction.

What is fixable, even on old hardware, is everything competing with that processing for the same limited resources — and that's usually most of the actual gap.

Working through the fixable causes, in order

Memory pressure from other open apps. Local processing needs a share of the Mac's memory to work with, and a browser with dozens of tabs, a video call, or a heavy design or video app running at the same time leaves less of it available — the effect is more pronounced on a Mac that had less memory to begin with when it shipped. Open Activity Monitor, look at the Memory tab, and check whether memory pressure is showing yellow or red rather than green while dictation feels slow.

A background task using the processor. A Time Machine backup, a Spotlight re-index after a large file operation, or a software update downloading in the background all compete for the same processor that speech processing needs. Activity Monitor's CPU tab, sorted by usage, usually makes an obvious background hog visible within a few seconds of looking.

Thermal throttling on a laptop. Older laptops, especially ones used propped on a soft surface or with a blocked vent, can throttle their own processor speed when they run hot, and a Mac that has been doing something demanding just before a dictation may still be in that reduced-speed state when the dictation happens. Letting it cool for a minute, or moving it to a hard, ventilated surface, is a real fix for this specific case, not a placebo.

A nearly full startup disk. macOS uses free disk space as working memory when physical memory runs short, and a drive with only a few gigabytes free gives it very little room to do that. Checking free space in About This Mac → Storage and clearing some if it's close to full is worth doing before assuming the processor is the bottleneck.

Ruling out length as the actual explanation

Before troubleshooting hardware at all, it's worth confirming the "slow" dictation wasn't simply a long one. A sentence takes a fraction of a second; two minutes of continuous speech takes meaningfully longer to turn into text, on any Mac, old or new — that scaling is expected and not evidence of a problem. If the wait tracks with how much was said rather than showing up even on short sentences, this is that, not a hardware issue.

What doesn't help, and why

Restarting the app rarely changes anything here, because the slowdown isn't a stuck process — it's genuinely how long local processing takes given what else the Mac is doing at that moment. The one exception is a Mac that has been awake for a very long time without a restart, where background processes can accumulate in a way a restart does clear; if nothing else on this page explains it, that's worth trying last, not first.

Turning off other Coii VoiceInput features doesn't meaningfully speed this up either — filler removal and dictionary lookups are lightweight text operations that run after the slower step, not instead of it, so disabling them saves very little.

What this doesn't mean for privacy or where the audio goes

None of the above involves the network, and that's worth being explicit about: nothing about how dictation is processed sends audio anywhere, regardless of how long it takes locally. A slow dictation on a bad day is still a dictation that never left the Mac, which is a different property from being fast, and worth not confusing with a network problem just because both can involve waiting.

If a slow dictation is also going missing

Slowness and a dictation not appearing at all are different problems with different causes, and it's worth not conflating them. What happens to a dictation that doesn't land where expected covers the second case specifically — the short version is that the transcript is kept in history regardless of how long processing took or where the result ended up, so a slow dictation that also seems to have vanished is almost always sitting there, not lost.

Checking whether the pattern is constant or occasional

It's worth noting whether the slowness happens on every single dictation or only sometimes, because the two point in different directions. A wait that's the same length every time, regardless of what else is running, points toward the hardware-ceiling explanation above — the Mac is doing about as well as it can, consistently. A wait that varies from one dictation to the next, fast sometimes and slow other times with nothing obviously different about what was said, points toward one of the fixable causes — background tasks, memory pressure, or heat — since those come and go depending on what else the Mac happens to be doing at that particular moment. Keeping a rough mental note across a handful of dictations, rather than judging from a single slow one, makes it much easier to tell which situation is actually happening.

A second Mac, side by side, as a diagnostic

If a second, newer Mac is available, the single most reliable way to separate "this Mac is old" from "something fixable is happening right now" is to try the same short dictation on both, ideally with nothing else open on either one. A small, consistent gap between the two that stays roughly the same across several tries is the hardware difference talked about above, and not something to keep chasing with settings changes. A gap that's large, inconsistent, or that shrinks once background apps are closed on the older Mac points the other way — toward something fixable on that specific machine at that specific moment, rather than a permanent property of its age.

Setting realistic expectations on the oldest supported Macs

For a Mac at the very bottom of what macOS 13 Ventura supports, a consistently longer wait than half a second, even with nothing else running and plenty of free memory, is a real property of that hardware rather than a problem with a fix. Apple's own built-in dictation has the advantage of being tuned specifically to Apple's own hardware across its full range, which is one of the places it can have an edge on genuinely old machines — worth knowing honestly rather than chasing a setting that isn't there. On hardware in that position, the fixable causes above — memory pressure, background tasks, heat, free disk space — are still worth ruling out first, since they're common enough to explain most of the gap even on an older machine, before accepting the remainder as the hardware's own ceiling.

Try it and judge the speed against your own Mac's actual load, not a single dictation taken in isolation — the difference between a Mac doing nothing else and a Mac running a backup in the background is often larger than the difference between two Macs several years apart.

A short routine worth running once, rather than guessing repeatedly

Rather than adjusting one setting at a time and redictating after each change, it's faster overall to gather all the relevant information first: open Activity Monitor and note the memory pressure color and the top processor-using background task, if any; check free disk space in About This Mac; and try one short dictation and one longer one, timing each roughly by feel rather than precisely. With that in hand, the cause is usually obvious — a red or yellow memory indicator, a backup or index visibly running, a nearly full drive, or none of those, in which case the remaining explanation is the hardware ceiling described above rather than anything fixable in the moment. This one pass through the checklist tends to answer the question more conclusively than trying fixes individually and re-testing after each one, which is slower and easier to draw the wrong conclusion from if two causes happen to be present at once.

Why this app runs the processing locally at all, given the tradeoff

It's a fair question, on an older Mac, why the tradeoff is worth it: the alternative most other dictation tools take is sending audio to a server and getting text back over the network, which sidesteps the local-hardware ceiling entirely at the cost of every recording leaving the machine. That tradeoff runs the other way for anyone who needs dictation to work with the network off, on a locked-down machine, or with nothing about what was said ever transmitted anywhere — for those cases, a slower result on old hardware is a real cost, but it's a cost paid in exchange for a guarantee a networked tool structurally cannot make, not a shortcoming with no upside attached to it.