Dictation Software: How to Choose the Right App

Most people should start with the free dictation already built into their Mac, Windows PC, phone or browser, and pay for software only when a daily task proves it falls short. Dictation is worth the effort: in one controlled test, speech entered English text on an iPhone at 153 words per minute, against 52 on the built-in keyboard ^{1 2}. The choice of tool then comes down to four questions. Is your voice processed on your device or in the cloud? Do you want your raw words or AI rewriting? Do you prefer a subscription or a one-time license? Do you need a general tool or a specialist one? A ten-minute test on your own writing settles the rest better than any ranking, including ours.

This page maps the whole subject and links to a guide on each part. Yap, which we make, is an on-device dictation app for Mac and Windows. It appears only where it belongs in a fair comparison, and the page recommends the built-in tools whenever they are enough. Product facts below were checked on 2026-10-16 and will change, so treat prices and limits as a dated snapshot.

Dictation software types what you say into the app you are using

Dictation software turns speech into text at your cursor, in the document, email or chat you already have open. It is different from a transcription tool, which converts a finished recording into a file, and from voice control, which operates the computer with commands. Many products do two of the three, and that blur is why searches for "the best dictation app" return such mixed lists. The full definition, with examples of each kind, is in our guide to what dictation software is. Short answers to the most common questions sit in separate pages: what voice recognition software is used for, whether voice recognition is the same as speaker recognition and the difference between speech to text and text to speech.

Modern products run in two steps. A speech recognizer turns audio into words, and an optional second step cleans them up. Open models show how far the first step has come: OpenAI's Whisper models were trained on 680,000 hours of audio with transcripts ³, and their authors report that the models approach human accuracy and robustness ⁴. The second step is where "AI dictation" differs from older software, as the section on AI apps below explains. If you want the technology rather than the products, see our overview of speech recognition.

Six kinds of software, sorted by where your voice goes

The categories matter because each one fixes a different trade-off between price, privacy, coverage and polish.

Built-in dictation comes first because it costs nothing and is already installed. The platform details differ, and they decide whether you ever need anything else.

Built-in dictation is free and covers more than most people expect

Every major platform ships dictation. Apple Dictation on a Mac lets you speak to enter text anywhere you can type it ⁵, and Apple says that on a Mac, voice inputs and transcripts for general text Dictation are processed on your device and not sent to Siri servers ⁶. You can check this yourself, because Keyboard settings show whether dictated audio is handled on the device or sent to Apple ⁷. Apple's macOS availability page lists 63 language and regional variants for Dictation on the Mac ⁸. iPhone dictation is processed on the device in many languages with no internet connection required ⁹.

Windows voice typing works differently. Microsoft says it uses online speech recognition powered by Azure Speech services ¹⁰, and it needs a working microphone, an internet connection and the cursor in a text box ¹¹. If you need hands-free control without a connection, Windows 11 Voice Access controls the PC and writes text by voice without an internet connection ¹², although its first-time setup needs one ¹³. It is available on Windows 11, version 22H2 or later ¹⁴. Microsoft says the older Windows speech recognition is no longer being developed and is being replaced by Voice Access ¹⁵.

On Android, you tap the microphone at the top of Gboard and speak ¹⁶, and Google moved Gboard voice typing onto the phone in 2019 with an 80 MB recognizer that works offline, first on Pixel phones in American English ¹⁷. In the browser, Google Docs voice typing runs from Tools, then Voice typing, in a supported browser ¹⁸, and Google says it works with the latest Chrome, Edge and Safari ¹⁹. Our free online dictation page also works in the browser without an account.

The detail is in our guide to free dictation software. Four short answers cover the big vendors: does Apple have a voice to text app, does Google have one, does Microsoft have dictation software and how much dictation software costs. Platform-specific depth lives in our overviews of dictation on Mac, dictation on Windows and dictation on phones and tablets.

Where built-in dictation stops

Built-in tools have limits you can feel within a day. They rarely clean up filler words, and they do not rewrite for tone. Mac Dictation stops by itself when it hears no speech for 30 seconds ²⁰, and on a Mac it will mark ambiguous text with a blue underline for you to resolve ²¹. Windows has started to close the gap: its fluid dictation corrects grammar, punctuation and filler words as you speak ²². Platform differences also appear in what the apps need: Word's Dictate feature works in Microsoft 365 and needs a reliable internet connection ²³. These are the gaps a paid app has to fill to be worth the money.

Pay for every-app typing, offline privacy, AI cleanup or a specialist vocabulary

There are five reasons to pay, and if none applies to you, keep the free tool.

1. Coverage. The tool must type into every program you use, not just the ones with a built-in microphone button.
2. Offline and private processing. Audio must stay on your machine, and you must be able to verify that.
3. AI cleanup. You want fillers removed, punctuation fixed and sentences tightened without editing by hand.
4. A specialist vocabulary. You dictate medical, legal or technical terms that general models miss.
5. Full hands-free control. You need to operate the computer, not only enter text.

The pricing models follow from these needs. Most AI apps use a free tier with a cap, then a subscription. A few sell a perpetual license, paid once. Open-source tools cost nothing. The guide to free dictation software covers the free end in detail, and the question on how much dictation software costs gives the short version.

Free tiers cap your words or switch off the AI

Free tiers of paid apps are generous in different ways, and the caps decide whether they are usable for real writing. Wispr Flow's free plan allows 2,000 dictated words per week on desktop and 1,000 per week on mobile ²⁴, and its help center says the desktop count is a combined total of desktop and mobile dictation ²⁵. Aqua Voice's free plan includes 1,000 free words ²⁶. Willow Voice's free plan includes an unlimited weaker speech-to-text model and 20 Scribe uses per week ²⁷. Superwhisper's free plan takes the opposite approach: unlimited dictation with local Whisper models, but without AI processing ²⁸.

For a reader who writes a few thousand words a week, a 2,000-word weekly cap ²⁴ is a trial, not a free plan. A reader who dictates only short messages may never hit it. Counting your own weekly words before you compare tiers is the single most useful step, and our question on how to test dictation software shows how.

Subscription or one-time license: the arithmetic

Pricing models differ more than the headline numbers suggest. The chart compares the cheapest listed way to buy a year or more of each tool, using prices checked on 2026-10-16. Wispr Flow is shown at its listed Pro price of $15 per month ²⁹, which comes to $180 over a year ³⁰. Dragon Professional 16 is shown at the $685.99 one-user license price from one reseller, CDW ³¹, not a vendor list price, so treat it as one data point.

VoiceInk's Solo license costs $25 once, for one Mac, with lifetime access ³². Superwhisper lists its Pro plan at $84.99 per year ³³ or $249.99 as a one-time lifetime purchase ³⁴, and one license works on unlimited Macs and Windows PCs plus an iPhone and an Android phone ³⁵. At its monthly price of $8.49 ³⁶, the lifetime license costs the same as about 29 months of subscription ³⁷. That break-even is a rule of thumb: at the monthly price, a one-time license wins if you keep using the tool for longer than about 29 months ³⁷, provided the vendor stays in business. Subscriptions win for people who switch tools often or want the vendor to keep paying for cloud models.

The leading apps differ on processing, platforms, price and how much they rewrite

The apps covered in this section are Wispr Flow, Aqua Voice, Superwhisper, MacWhisper, VoiceInk, Willow and Otter. They are less alike than their marketing suggests, and the four axes that separate them are where audio is processed, which platforms they run on, how they charge and how much they change your words. Our guide to AI dictation apps compared goes through each with dated vendor facts. This is the short version.

Two details deserve emphasis. First, cloud and local are not marketing words but a real difference in who holds your audio. A cloud app sends audio to a server, which the vendor may or may not store, and a local app does not. Wispr Flow's privacy page says it deletes audio once transcribed unless you enable cloud storage ³⁸, while Aqua Voice's privacy policy says it may store transcript data for users with Privacy Mode off ⁴². Neither is a defect, but the settings differ, so read them. Second, rewriting is a spectrum. A free local Whisper setup gives you words as spoken, with punctuation. A cloud app with a language model removes fillers, restructures sentences and can match a tone. In one benchmark, a large language model could correct words missing from the recognizer's candidate list, which the researchers attribute to its ability to generate text rather than only choose among candidates ⁵². The flip side is that rewriting can change what you meant, and some people want the raw words. There is a short answer on dictation apps without AI.

Head-to-head comparisons on this site cover specific pairs: Wispr Flow alternatives, Aqua Voice alternatives, MacWhisper for Windows and VoiceInk for Windows. Questions on individual apps include whether Wispr Flow is worth it, Wispr Flow versus Superwhisper and whether Superwhisper beats Apple Dictation. For notes and journaling, see the best voice to text app for notes.

Meeting transcribers solve a different problem

Otter and similar tools join a meeting, record it and produce a transcript with speakers and summaries. That is a transcription job, not dictation: nothing is typed at your cursor as you speak. If your goal is notes from calls, a transcription suite is the right category. If your goal is writing emails and documents by voice, it is not. Our overview of transcription covers that side.

Dragon defined the category, and it is now a narrow Windows specialist tool

Dragon NaturallySpeaking is the best-known name in dictation. In April 1997 Dragon Systems unveiled it as the first general-purpose, large-vocabulary continuous speech recognition product ⁵³. Ownership then moved: a court-approved sale in November 2001 of Lernout and Hauspie's speech assets included the Dragon line, which was to go to ScanSoft ⁵⁴, and Microsoft announced on March 4, 2022 that it had completed its acquisition of Nuance ⁵⁵. The timeline and what each owner changed are in our guide to Dragon NaturallySpeaking, with short answers on whether it is discontinued, whether Microsoft owns it, whether it uses AI and whether it is worth it.

What still exists, and what has ended

The Mac version is gone. Nuance said it had decided to discontinue the Dragon Professional Individual for Mac line ⁵⁶, and updates ended after 22 October 2018 ⁵⁷. Nuance offered telephone support for the discontinued Mac product for up to 90 days from activation in North America ⁵⁸. Consumer editions went the same way: Nuance withdrew Dragon Premium 13 from the market effective September 1, 2018 ⁵⁹, and a reseller reports that Dragon Home is no longer available for sale ⁶⁰. Nuance's US store menu lists Dragon Legal, Dragon Law Enforcement, Dragon Professional Anywhere and Dragon Legal Anywhere, with no Home or Premium ⁶¹. If you are on a Mac, see our page on Dragon for Mac rather than hunting for an old installer.

The current desktop product is Dragon Professional 16 for Windows. Nuance lists Windows 11 (64-bit), Windows 10 and several Windows Server versions as supported ⁶², and the program needs 4 GB of RAM and 12 GB of free disk space ^{63 64}. Its engine is powered by Nuance Deep Learning technology ⁶⁵. A reseller listed a one-user license at $685.99 ³¹, which is about 2.29 times the $300 launch download price of Dragon Professional Individual 15 in 2016, though the two are different editions ⁶⁶. A cloud variant, Dragon Professional Anywhere, was listed by one reseller at $660 per year ⁶⁷, another at $480 ⁶⁸, so check the current price with the vendor. All version, edition and price detail is in the guide to Dragon NaturallySpeaking versions, editions and price. Short answers include how much Dragon costs, Dragon Home versus Professional, Dragon 15 versus 16, whether Dragon is free, whether there is a cloud version and its system requirements.

Where Dragon still wins

Dragon is not obsolete. It leads in three areas. Voice commands are a strength: MouseGrid divides your screen or active window into nine numbered areas so you can move the pointer by number ⁶⁹, on up to three monitors ⁷⁰. Custom vocabulary and training are mature: you add a word with its spoken form ⁷¹, and your user profile stores your microphone type and custom words ⁷². And the standard installation runs locally on your PC. It also needs care: Nuance's own guidance says Dragon records everything it hears unless you tell it to stop listening ⁷³.

The newer apps win on setup, price and formatting. In a 2025 pilot of three physicians ⁷⁴, a Whisper-based model had a mean word error rate of 1.75 percent, against 1.54 percent for Dragon Medical One, and the difference was not statistically significant ^{75 76 77}. It is a small study, and it argues that accuracy alone no longer separates the two. If you still want to learn Dragon, the guide on how to use Dragon NaturallySpeaking covers setup and commands, with answers on commands, moving it to a new computer, when it stops working, microphones, the web extension and JAWS compatibility.

Replace Dragon by the job you used it for

If you used Dragon and need to move on, the right replacement depends on which job you relied on. Plain dictation is now handled well by free built-in tools and by the apps above. Automatic formatting is the strength of AI apps. Hands-free control is the hardest to replace. On Mac, Apple Voice Control handles it, and turning it on means it is used for dictation and standard macOS Dictation is not available ⁷⁸. On Windows, Voice Access fills the same role. Talon, a hands-free tool for writing, coding, browsing and gaming ⁷⁹, offers downloads for macOS, Windows and Linux ⁸⁰.

Before you switch, copy out your custom words and commands, because they are the hardest part to recreate. The guide to Dragon NaturallySpeaking alternatives walks through each replacement, and three short answers cover the common cases: a free alternative to Dragon, whether AI dictation beats Dragon and alternatives that also control the computer by voice. If you rely on voice control for access reasons, our overview of accessibility and dictation goes further.

Open-source dictation is free with no word limit, and you can read the code

Open-source dictation apps run open speech models on your computer, cost nothing or little and let anyone audit what happens to your voice. The active projects include Handy, VoiceInk, OpenWhispr and Hex. Handy describes itself as a free, open source speech-to-text application that works completely offline ⁸¹, released under the MIT License ⁸², and works on Windows, macOS and Linux ⁸³. On 16 October 2026 its releases page marked v0.9.8 as the latest release ⁸⁴. OpenWhispr is a cross-platform dictation app for macOS, Windows and Linux ⁸⁵, and its README states that it has no data collection and no telemetry ⁸⁶. HEX describes itself as local-first voice dictation for Apple silicon Macs running macOS 15 or newer ⁸⁷. Separately, the original Swift Hex repository was archived by its owner on 14 October 2026 and is now read-only ⁸⁸, so check which version you are installing. That is the kind of detail a lot of lists never date.

Open source is not the same as private by default. The HEX README says that when post-processing with OpenCode is turned on in Modes, the app sends the transcript, the foreground application and the Brave website hostname to the configured model provider ⁸⁹. The privacy benefit is that you can read the code and the settings, not that every feature stays local.

The underlying models are open too. Whisper's code and model weights use the MIT License ⁹⁰, and the smallest model has 39 million parameters and needs about 1 GB of memory ⁹¹, while the largest has 1,550 million parameters and needs about 10 GB ⁹². NVIDIA's Parakeet v3 model has 600 million parameters ⁹³ and extends support from English to 25 European languages ⁹⁴. In practice this means a recent laptop can run a model small enough for fast dictation, and a stronger machine can run a bigger one for better accuracy.

The trade-offs are setup time, fewer integrations and volunteer maintenance. The guide to open-source dictation software compares projects by license, platform, model and last release, and explains how to judge a project before installing it. Short answers cover the Handy dictation app, the Hex dictation app and whether VoiceInk is open source. For the privacy side of local versus cloud, see our overview of voice privacy, and for offline use on a Mac see our page on offline dictation for Mac.

Specialist fields need specialist dictation software

Doctors, radiologists, lawyers, police officers and vets often need more than a general app. A specialist product adds a field vocabulary, connects to the records system and carries the compliance terms an employer requires. Epic's mobile apps are a good example of the integration issue: a health system's tips note that Epic's Haiku (smartphones) and Canto (iPads) apps support Nuance Dragon dictation directly into progress notes ⁹⁵, and that Dragon Medical One is the tool for dictating into a patient's chart on a desktop or tablet in Hyperspace ⁹⁶. Microsoft's Dragon Copilot combines and extends Dragon Medical One and DAX Copilot ⁹⁷. In radiology, a 2007 report said Yale-New Haven Hospital's radiology department planned to use PowerScribe to cut turnaround for more than 365,000 reports a year ⁹⁸, and was planned to integrate with its records and imaging systems ⁹⁹.

The guide to professional dictation software compares these products by field. It also separates three things that get lumped together: dictation, digital dictation platforms that route recordings from author to typist, and ambient scribes that write notes from a conversation. Short answers include whether Epic has dictation, what radiologists use, what Heidi is, veterinary dictation, police report dictation and Philips and Olympus recorder software. How professions actually dictate is the subject of our overview of dictation at work.

A general app is enough when nothing you dictate has to reach a records system. A writer, a student or a manager rarely needs a specialist license. For recommendations by role, see the questions on dictation software for writers and dictation apps for kids. If a child will use the tool, note that the FTC's amended children's privacy rule adds biometric identifiers that can recognize a person to its definition of personal information ¹⁰⁰.

Accuracy depends on your microphone and room more than on the brand

Every app claims high accuracy, and none of the claims tell you how it will do on your voice. The size of the effect is large. In one study using deep neural network models, close-talking Wall Street Journal sentences recorded in a studio had a word error rate of 3.7% ¹⁰¹, and a single distant ceiling microphone in an apartment gave 12.0% ¹⁰².

Word error rate is the standard score for this comparison. Researchers describe it as a standard measure of the discrepancy between machine and human transcriptions ¹⁰³. Accuracy also varies across speakers: across five commercial systems, the average word error rate was 0.35 for Black speakers and 0.19 for white speakers ¹⁰⁴. So a vendor's headline figure on its own test audio tells you little about your own accent and room. Nuance's guidance for Dragon puts a headset microphone about 1 inch from the side of the mouth, not touching it ¹⁰⁵, and Apple advises that a headset microphone may help in a noisy or echoing room ¹⁰⁶. For most people, a closer microphone is the cheapest improvement.

For the full comparison of accuracy claims see the guide to best dictation software and the question on the most accurate dictation software.

Test any dictation app on your own writing before you pay

A fair test takes about ten minutes and beats any review. The longer version is in the question on how to test dictation software before buying, and the steps are these.

1. Pick two passages from your real work, about 150 words each: one plain, one with names, numbers or jargon you actually use.
2. Use the microphone you will use every day, in the room where you will use it.
3. Dictate each passage into the free built-in tool and into the app, reading naturally at a conversational pace, as Nuance advises ¹⁰⁷, and in whole phrases rather than single words ¹⁰⁸.
4. Count the wrong words, and note the minutes it takes to fix each draft. The time to correct matters more than the raw error count.
5. Check where the audio goes in the app's privacy settings, and whether it works with no connection if you need that.
6. Check what the free tier limits and whether the paid price fits how long you will use the tool.

Speech remains faster than typing even after corrections, but editing time is the cost. In the iPhone study, speech had a lower corrected error rate during entry than the keyboard (5.30% against 11.22%) but left a slightly higher uncorrected error rate in the final text (1.30% against 0.79%) ¹⁰⁹. The authors also say that results outside the laboratory need more study ¹¹⁰. Your own test is the evidence that applies to you. Habits matter too: users in a longitudinal study became more efficient at dictation, navigation and error correction as they gained experience ¹¹¹, so give a tool a week before judging it. Our overview of dictation covers how to build the habit, and a short blog post on choosing a dictation app gives a one-page checklist.

A decision table by situation

If you have read this far and want the answer in one place, use this table. It assumes you have already tried the built-in tool and it fell short.

The thesis holds across every row: free first, and pay for the specific thing the free tool cannot do. To go deeper on any row, start from the guide named in the section above, or read the guides on what dictation software is, the best dictation software by need and free dictation software. To try Yap on a Mac or Windows PC, see download and pricing; it is one option among those above, not a prerequisite for anything on this page.

Related subjects

• Dictation: how to dictate well, commands, punctuation and habits.
• Dictation on Mac and on Windows: platform-specific tools and fixes.
• Dictation on phones and tablets.
• Speech recognition: how the models work and how accuracy is measured.
• Transcription: turning recordings into text.
• Voice privacy: where your audio goes and who can keep it.
• Dictation at work and accessibility: professional and assistive use.
• Typing speed: how dictation compares with the keyboard.