The tools built for a sentence are not the tools built for a chapter

Most dictation software is tuned, whether its makers say so or not, around the shortest common case: a text message, a Slack reply, a two-line email. That is the dictation almost everyone does most often, so it is a reasonable place to optimise. It is also a different job from opening a blank document and talking for forty minutes until a full first draft of a report, a chapter or a long client email exists. Punctuation has to hold up over paragraphs rather than one sentence. A stalled transcription costs you a whole train of thought instead of one line. And the app needs to stay out of the way for long enough that you forget it is there. This page is about six Mac tools that people actually use for that longer job, what each one does well across a real hour of speech, and what happens when something goes wrong partway through.

What actually changes over a long session

Three things separate a good short-dictation tool from a good long-form one, and none of them show up in a thirty-second demo. Punctuation has to survive without constant correction — a tool that gets commas right in one sentence but drifts over ten paragraphs is a tool you end up editing more than typing would have cost. A failure partway through has to be recoverable — the longer the session, the more expensive it is to lose the last few minutes to a crashed engine or a network drop. The app has to disappear — a dictation tool that needs re-triggering, re-focusing, or babysitting every few sentences turns a chapter into a hundred small interruptions.

The alternatives

Wispr Flow — a vocabulary that keeps learning

Wispr Flow's pricing page states Pro includes a vocabulary that learns your names and jargon and syncs it across every device you use, at $15 a month (20% off billed annually), with a free tier available and Mac, Windows, iOS and Android support. For someone dictating the same project names and technical terms daily across a laptop and a phone, that learned, synced vocabulary is a genuine advantage this app does not attempt to match — Coii VoiceInput's corrections stay on the one Mac they were made on and do not train anything. The comparison covers the rest of a considerably more capable product on most other axes.

MacWhisper — cleans up the prose after you stop talking

Checked on macwhisper.com today: the free tier does core transcription, and Pro, a €64 one-time purchase, adds system-wide dictation plus "Automatic grammar improvement" and the ability to "Dictate with AI prompts" — a genuinely useful pass for someone who talks in run-on sentences and wants the prose tightened before it is called a draft. MacWhisper is also, at its core, a file-transcription app, so a chapter you already recorded as a voice memo on a walk can be turned into text after the fact, which none of the live-dictation tools on this page can do at all. The comparison has the rest.

superwhisper — local, with a trained vocabulary of its own

Checked on superwhisper.com today: Pro's feature table lists "Custom vocabulary" and "Custom modes" as paid capabilities, on top of the local transcription every tier gets, at $8.49 a month, $84.99 a year, or $249.99 once, per their comparison page. The custom modes are worth noting specifically for long-form use: a mode can be tuned for a particular kind of writing, which a fixed single-purpose tool like this one does not offer. The comparison covers the rest of a much larger product.

VoiceInk — one purchase, no monthly meter running

Checked on tryvoiceink.com today: Solo is $25 for one Mac, Personal $39 for two, Extended $49 for three, all one-time with lifetime updates and a 14-day money-back guarantee. It is open source, which means the app's actual behaviour during a long session — what it does with your audio, what it stores — is something you or anyone else can go and read rather than take on trust. The comparison covers the rest of a genuinely capable app.

Dragon — the category's command-and-macro leader, not available here

Nuance's own support page confirms Dragon Professional Individual for Mac 6 was "discontinued effective 10/22/2018", and the current Dragon Professional product page is labelled Windows, "optimized for Windows 11." It earns a place on this page anyway because a two-decade professional dictation market grew up around exactly the long-form problem this page is about, and Dragon's custom voice commands and multi-step macros are still the feature set every other tool is measured against. A Mac user cannot install it, which is worth saying plainly rather than pretending the comparison is live. The comparison has the full inventory.

Apple Dictation — free, and fine for shorter pulls

Free, already on the Mac, and it needs nothing installed. What it does not have is any of the three things this page opened with: no vocabulary that learns your terms, no published guarantee about what happens if a connection drops mid-session, and no editing-by-command layer — it types what you said and stops. For someone whose "long-form" is a few paragraphs at a time rather than a sustained hour, it is a perfectly reasonable place to start before paying for anything. The comparison covers the rest.

Talon — for when the real bottleneck is editing, not composing

Talon is not a dictation app and doesn't try to be one — it is a free, downloadable control layer for macOS, Linux and Windows that lets you select, delete, move and rewrite existing text by voice command, on top of dictating new prose. If your long-form sessions are mostly revision — restructuring a draft that already exists rather than producing one from nothing — that editing vocabulary is the actual bottleneck, and none of the other tools on this page, including this one, address it. The comparison covers what it does and does not do.

Coii VoiceInput — one key, and a sentence that cannot go missing

$19 once, three devices, macOS 13 or later. No vocabulary that learns your terms across sessions, no editing commands, no file transcription — it is a narrower tool than most of the list above. What it commits to specifically is the failure case a long session actually risks: every sentence lands in your local history the moment you release the key, before the engine has produced anything, so a slow model or a crashed engine in minute thirty costs you nothing already said. Nothing is uploaded, and it works with the network off.

The table

What none of these fully solve

None of these apps, including this one, does the actual writing for you — a long dictated draft still needs the same structural editing a long typed draft would, and speech tends to produce more repetition and fewer paragraph breaks than typing does, simply because talking doesn't pause the way a keyboard does between thoughts. MacWhisper's grammar pass and superwhisper's custom modes both narrow that gap somewhat; none of the others, this one included, touch it at all. Planning the shape of what you're about to say before you start talking remains the most effective fix regardless of which app is listening.

What actually carries across if you switch

Nothing here locks you in. None of these apps store your drafts in a format only they can read — the output in every case is plain text, typed into whatever document you were already writing in your own choice of editor. The only thing that does not carry across is a learned vocabulary: Wispr Flow's and superwhisper's adapt over weeks of use, and that adaptation resets if you switch away from them. If your terms are unusual enough that this matters, that's the real cost of trying something else, not any file format or export step.

What a real hour looks like, tool by tool

With Coii VoiceInput, an hour of drafting is a series of held keypresses, each one landing in history the instant it ends — the session has no state beyond that: no mode to remember is on, nothing to forget to turn off before a call. With Wispr Flow or superwhisper, the same hour benefits from a vocabulary that has already learned your recurring names and terms, so correction time drops as the session goes on rather than staying flat. With MacWhisper, the same hour might not be live dictation at all — a voice memo recorded on a walk, imported afterward and cleaned up with a grammar pass, is a genuinely different shape of workflow, and one none of the live-only tools here support. With Talon or Dragon, the hour includes revision as well as composition: selecting a paragraph and rewriting it by voice rather than retyping it, which changes the total time spent far more than any transcription-accuracy difference between tools would.

The punctuation problem specifically

Punctuation is where short-dictation tools tend to show their seams first over a long session. A comma or a period placed correctly in one sentence is a much easier problem than consistent punctuation across fifteen minutes of continuous, unstructured speech, where pauses for thought don't map cleanly onto sentence boundaries. None of the vendors on this page publish a punctuation-accuracy figure specific to long sessions, so the honest approach is to test with your own real material — read a paragraph you actually intend to write, at your normal pace including the pauses, and look at what came out before trusting any tool with something that matters.

Reading it back matters more here than anywhere else

The cost of an uncaught error scales with length. A wrong word in a one-line Slack message is a quick fix; a wrong word or a garbled clause buried in paragraph twelve of a report is easy to miss and expensive if it isn't. That argues for reading back long-form dictation more carefully than short dictation, regardless of which tool produced it — none of the apps on this page, including this one, check facts or catch a misheard number, and a long document gives more places for exactly that to hide unnoticed.

The interruption a long session actually risks

A short dictation that fails costs a few seconds to redo. A long one that fails costs whatever wasn't yet saved — a real distinction that only shows up once a session runs long enough for something to plausibly go wrong partway through: a laptop moving to sleep, a Wi-Fi drop for a cloud-dependent tool, an engine that stalls on an unusually long utterance. None of the vendors on this page other than this one publish a specific guarantee about what happens to the words already spoken when that failure occurs mid-session, which is worth weighing against the other features here if losing thirty minutes of drafting to a stall is the actual risk you're trying to avoid.

Picking one

If the actual friction is that punctuation and editing commands would let you revise a draft entirely by voice, look at Talon, or at Dragon if you can run Windows. If you want a vocabulary that keeps learning your terms across a laptop and a phone, Wispr Flow is built for exactly that, for $15 a month. If your real source material is often a recorded voice memo rather than live speech, MacWhisper's file transcription plus its grammar pass is the more direct tool. If none of that applies and the actual requirement is simpler — open the app, talk for as long as it takes, and know that nothing already said can go missing along the way — the trial covers that for thirty days before the $19 is ever charged.