Speech-to-text for accountants: talk three times faster than you type

Here's a number worth sitting with. The average professional types around 40 to 60 words per minute. The average person speaks at 130 to 160. A widely cited Stanford study found speech input roughly three times faster than typing, with a lower error rate. That study measured phone keyboards, so treat the exact multiple with care on a full keyboard, but the direction isn't in dispute: your mouth is much faster than your fingers.

Now think about what an accountant's day actually looks like. Emails to clients. File notes after calls. Review points on a junior's workpapers. Meeting prep. Almost none of it is data entry. Most of it is composing sentences, which is exactly the work dictation is built for.

The honest version of the 3x claim: the raw speed gain is real, and the bigger win is that drafting out loud removes the composing tax. When you type, you edit as you go, second-guess phrasing, and stall on openings. When you talk, you explain the thing the way you'd explain it to a colleague, and modern tools clean it up as you speak. People who adopt dictation rarely rave about words per minute. They rave about emails that used to sit in drafts for an hour going out in two minutes.

Do this now: count how many emails and file notes you wrote yesterday. If it's more than ten, dictation is worth an afternoon of your time.

How modern AI dictation actually works

If your reference point is Dragon NaturallySpeaking circa 2010, delete it. Old dictation transcribed literally: you said "um, so, comma, new paragraph" and prayed. Modern tools run your speech through a large speech-recognition model (often a descendant of OpenAI's open-source Whisper), then pass the raw transcript through a language model that punctuates it, strips the ums and false starts, and formats the result as clean prose. They sit at the system level, so they work in every app: Outlook, Xero, FYI, Teams, your browser. Hold a hotkey, talk, release, and finished text appears wherever your cursor is. No voice training, no vocabulary files, no separate dictation window. As we argued in our guide to AI for accounting firms, the models finally got good enough that the tooling disappeared into the background.

Do this now: if you're on a Mac or Windows 11 machine, you already have free dictation installed. Try it on your next email before spending a cent (details below).

The main options compared

Pricing below is as at July 2026 and worth re-checking before you buy, because this market moves quickly.

Tool	Price (as at July 2026)	Platforms	On-device or cloud	Best for
Wispr Flow	Free tier (2,000 words/week); Pro about $12/month annual, $15/month monthly	Mac, Windows, iOS	Cloud	Most accountants: polished output, works everywhere
superwhisper	Free tier; about $8.49/month or $249 lifetime	Mac, iOS	On-device option (cloud modes available)	Privacy-conscious Mac users who want local processing
MacWhisper	Free tier; Pro around €59 one-off	Mac	On-device	Transcribing recordings and meeting audio, not live dictation
Aqua Voice	One-time 1,000-word free tier; about $8/month billed annually	Mac, Windows, iOS	Cloud	Fast real-time transcription with live text preview
Apple Dictation	Free	Mac, iPhone, iPad	On-device (Apple Silicon)	Free baseline, decent accuracy, no AI cleanup
Windows Voice Access	Free with Windows 11	Windows	On-device	Free baseline plus voice commands, fully offline
Talon	Free public version; paid beta via Patreon	Mac, Windows, Linux	On-device	Power users and full hands-free computer control

Tool-by-tool: honest verdicts

Wispr Flow

The current default recommendation, and the one most accountants should trial first. It runs everywhere, the AI cleanup is the best in the category, and it learns names and jargon over time (client names, "Div 7A", "franking credits") without manual vocabulary lists. Two caveats. It's cloud-only, so your audio leaves your machine every time you dictate (more on that below). And reviews are mixed on reliability, with a gap between strong editorial write-ups and weaker Trustpilot scores. Use the free tier for a week before paying.

superwhisper

The privacy pick. It can run its speech models entirely on-device, so client audio never leaves the laptop, and the $249 lifetime licence beats years of subscriptions. The trade-off: Mac-only, and setup involves more choices (which model, which mode) than Wispr Flow's install-and-go. For a firm with strict data policies and a Mac fleet, this is the sensible default.

MacWhisper

Different job. MacWhisper transcribes audio files (recorded meetings, voice memos, client calls) rather than live dictation into apps, all processed locally, and the roughly €59 one-off Pro licence is excellent value for that job. Buy it as a companion, not your daily driver. If meeting transcription is the actual problem, start with our guide to the best AI meeting assistants for accounting firms instead.

Aqua Voice

Fast and polished, with a live preview of your text as you speak, and the price undercuts Wispr Flow. The one-time 1,000-word free tier (it never resets) makes evaluation awkward: that's about eight minutes of talking, so plan your test before you start it.

Apple Dictation and Windows Voice Access

Genuinely usable free baselines, and both run offline on modern hardware, which quietly solves the confidentiality question. What you don't get is the AI cleanup layer: your ums, restarts and rambles land on the page as spoken. Start here, and upgrade once you know dictation sticks for you.

Talon

The deep end. Talon is a full on-device voice-control system, popular with programmers and people with RSI: you can dictate, but you can also drive every application entirely by voice. The learning curve runs to weeks. For an accountant with a wrist injury it's worth every hour; for everyone else it's more tool than the job needs.

Do this now: pick one paid trial and one free baseline, and run them head-to-head on real work for five business days. The right answer will be obvious by Wednesday.

Where it fits in a firm's day

Dictation earns its keep on the in-between writing that fills the gaps around actual accounting work:

• Client emails. The single biggest win. Explaining a tax position out loud takes two minutes; typing the same email takes ten, because you polish as you go. Dictate the substance, then spend thirty seconds tidying.
• File notes. The call ends, you hold the hotkey, and you narrate what was discussed and agreed while it's fresh. Ninety seconds of talking beats the five-line summary you'd have typed an hour later, and the reviewer (or the regulator) gets the detail. For client meetings specifically, an AI meeting assistant like Vinyl drafts the note for you before you reach the hotkey (Vinyl is a commercial partner of The Firm).
• Review points. Reviewing a junior's workpapers means writing a dozen short comments. Speaking them keeps you in the flow of the review instead of context-switching to compose each one. Bonus: spoken review points tend to come out kinder than typed ones.
• Meeting prep. Brain-dump the agenda, the client history and the three things you must not forget to raise, then let the transcript become your prep note. Pairs well with the prompt techniques in our AI prompts for accountants guide: dictate the messy version, let an AI structure it.

Do this now: tomorrow, dictate every file note instead of typing it. One day, one use case. That's the whole experiment.

The security questions to ask before your team dictates client info

Here's the part most reviews skip. A dictated file note about a client's restructure is client-confidential information, and with cloud tools it leaves your machine. Before rolling anything out firm-wide, answer four questions:

1. Where is the audio processed? On-device tools (superwhisper's local mode, MacWhisper, Apple Dictation on Apple Silicon, Windows Voice Access, Talon) process everything locally. Cloud tools (Wispr Flow, Aqua Voice, superwhisper's cloud modes) send audio to their servers, decrypted at their end for transcription.
2. Is anything retained, and is it used for training? Wispr Flow, for instance, offers a Privacy Mode plus settings it says amount to zero data retention, and holds SOC 2 Type II, ISO 27001 and HIPAA certifications. Reasonable posture, but defaults matter: check what's on out of the box, not what's possible in settings.
3. Does this fit your existing engagement terms? Your engagement letters probably already cover cloud processors (you use cloud accounting software, after all). Confirm dictation vendors fit the same framework rather than assuming.
4. What's the firm rule? Write one sentence of policy: which tool is approved, with which settings, and whether client-identifying dictation requires on-device processing. The thinking in our piece on whether AI is safe for client data applies here word for word.

Do this now: before anyone at your firm dictates a client name into a cloud tool, read that vendor's data retention page and screenshot the settings you've approved. Ten minutes now saves an awkward conversation later.

The foot pedal trick

Now for the power move. Every serious dictation tool is built around push-to-talk: hold a key, speak, release. That's deliberate, since push-to-talk beats always-listening for accuracy and for not transcribing your colleague's question across the desk. The culture around these tools is keyboard-shortcut culture: pick a hotkey you can hold comfortably (many people remap Caps Lock) and it becomes muscle memory within days.

The pedal takes it one step further. Buy any programmable USB foot pedal (generic single-pedal units are cheap, and transcriptionists have used them for decades), map it to your dictation hotkey with the vendor's configuration software, and your foot becomes the push-to-talk button. Your hands never leave the keyboard: cursor in the right cell of a workpaper, foot down, narrate the review point, foot up, keep moving. Honestly, it's a small gain over a well-chosen hotkey and most people don't need it, but for heavy review work, long file notes, or managing RSI it makes dictation feel less like an app and more like a reflex. Setup is unglamorous: the pedal presents itself as a keyboard, you assign it one key, done.

Do this now: master the hotkey first. If after a month you're dictating twenty times a day, order the pedal.

Frequently asked questions

Is dictation really three times faster than typing?

Roughly, for the drafting part. Speaking runs around 130 to 160 words per minute against typing speeds of 40 to 60, and Stanford research measured speech input at about three times keyboard speed on phones. On a full keyboard the multiple is smaller and you'll spend some time editing. Call it two to three times faster end to end for emails and notes.

What's the best free option?

Whatever's already on your machine. Apple Dictation runs on-device on Apple Silicon Macs, and Windows Voice Access ships free and fully offline with Windows 11. Neither cleans up your speech the way paid AI tools do, but both are good enough to prove whether dictation suits you. Wispr Flow's free tier (2,000 words a week as at July 2026) is the best way to sample the AI-cleanup experience.

Is it safe for client information?

On-device tools, yes: the audio never leaves your computer. Cloud tools require the same diligence as any processor of client data: check retention policies, training opt-outs and certifications, confirm the safe settings are actually enabled, and set a firm rule before rollout. When in doubt, keep client-identifying dictation on-device.

Does it work with accounting software?

Yes, and this is the big change from the old Dragon days. Modern dictation tools type into whatever field your cursor is in, so they work in Xero, Karbon, XPM, MYOB, Outlook, Teams and your browser exactly as if you'd typed. They won't navigate the software for you (that's voice control, which is Talon territory), but for any text field, they just work.

Won't I sound weird dictating in an open office?

For the first week, yes, a little. So did wearing headphones on calls once. Practical fixes: dictate right after calls end while the room's already used to you talking, batch dictation for the walk between meetings or the drive home via the mobile apps, and speak at normal phone-call volume, which is also what the models transcribe best. Most people stop noticing within days, including you.