What Is Dictation Software? Types and How It Works

Dictation software is a program that listens to your speech and types the words into whichever app has the cursor. It is built on automatic speech recognition (ASR), the technology that turns sound into words. Its job differs from a voice assistant's: it produces text you keep, such as an email, a report or a note. The dictionary sense of dictation is "the act or manner of uttering words to be transcribed" ¹.

Macs, Windows PCs, iPhones and many Android phones include a free version. Paid and specialist programs differ from it in where your voice is processed, what happens to the words afterward, and which vocabularies they handle. This guide sets out the six kinds of dictation software, what each does with your speech, and how the technology works from sound to text. For the wider subject, start with our overview of dictation software. If you are new to the word itself, see what dictation is. The question of what this technology is used for has its own answer: what voice recognition software is used for.

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

Dictation software is defined by where its text goes. It types your words where the cursor is: in a document, a message or a chat prompt. Apple says that with Dictation on a Mac you can speak to enter text anywhere you can type it ². Windows voice typing is similar, but it needs an internet connection, a working microphone and the cursor in a text box ³.

In macOS 27, Dictation accepts text of any length without a timeout, and you stop it with Escape, the Microphone key or the Dictation shortcut ⁴. System-wide dictation dates to 2012. Apple brought it to the Mac with OS X Mountain Lion, released on July 25, 2012 ⁵.

That typing behavior is what separates dictation from other voice tools. A dictation tool writes what you say. A voice assistant answers you or carries out a request. A voice control system operates the computer. A later section sets these jobs side by side.

Six kinds of dictation software differ in where your voice is processed

Two questions separate the kinds. The first is where recognition runs: on your device, on the vendor's servers, or on a server you control. The second is what the software does with the recognized words: insert them as they came out, or rewrite them with AI. The table gives the short version, and the subsections below check each row against sources.

Built-in tools run on the operating system, and their privacy depends on the platform

Apple says that on a Mac, general text Dictation is processed on your device and not sent to Siri servers, and you can check this in Keyboard settings ⁶. Apple also says Dictation on a Mac may need an internet connection ⁷, so the check in your settings matters. On iPhone, Apple says Dictation is processed on the device in many languages and needs no internet connection ⁸.

Windows voice typing works differently. Microsoft says it converts speech to text using online speech recognition ⁹. On Android, 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 ¹⁰.

For many people these free tools are enough. Our guide to free dictation software starts there.

Local-model apps such as Superwhisper can keep recognition on your computer

Superwhisper's free plan gives unlimited dictation with local Whisper models ¹¹. The company says nothing dictated leaves the device on that plan ¹². Its paid Pro plan adds cloud models and AI processing ¹³. That statement comes from the company about its own product, so treat it as a claim to check, not a measurement.

Cloud AI dictation apps send audio to servers and often rewrite the text

Wispr Flow lists dictation on Mac, Windows, iOS and Android ¹⁴. It says it deletes your audio as soon as it has been transcribed, unless you turn on its Dictation cloud storage setting ¹⁵. Aqua Voice says that for users with Privacy Mode turned off, it may store transcript data on its servers to improve the product ¹⁶. Apps marketed as AI dictation often fall in this group, and our comparison of AI dictation apps covers them in detail. Privacy settings differ from one app to the next, so read the policy for the one you choose.

Professional suites add vocabularies, user profiles and administration

Dragon NaturallySpeaking was unveiled in April 1997 as the first general-purpose, large-vocabulary continuous speech recognition product ¹⁷. Its professional successors add management features. Nuance says volume licensing and its Nuance Management Center let organizations deploy, manage and share custom words, commands and auto-texts ¹⁸. In Dragon for PC, you specify a word or phrase and its spoken form, and you can optionally train Dragon on its pronunciation ¹⁹. This is covered under custom vocabulary.

Clinical systems go further. In 2007, Yale-New Haven Hospital planned to use PowerScribe to cut turnaround for more than 365,000 reports a year ²⁰. Platform support can end. Nuance said it would no longer provide updates for Dragon Professional Individual for Mac after 22 October 2018 ²¹, so confirm that a product still supports your platform before you buy it.

Digital dictation workflow systems record audio for someone else to type

This kind predates speech recognition. A person records on a handheld device or microphone, and the audio goes to a transcriptionist or a recognizer later. The US Bureau of Labor Statistics says medical transcriptionists listen to the recorded dictation of physicians and other healthcare workers and transcribe it into medical reports ²². Philips' SpeechMike shows the hardware side: on a slide-switch model, you start recording by moving the switch to REC ²³.

Open-source tools let you run the models yourself

Handy describes itself as a free, open source speech-to-text application that works completely offline ²⁴. Its README says it is released under the MIT License ²⁵ and lists Windows, macOS and Linux ²⁶. HEX is local-first voice dictation for Apple silicon Macs running macOS 15 or newer ²⁷.

Open source gives you the code and control over the model, but the code alone does not show where your audio goes. HEX's README says that when post-processing is turned on in Modes, the app sends the transcript, the foreground application and, in Brave, the website hostname to the configured model provider ²⁸. Check each optional setting, not only the default.

Dictation, transcription and voice control are three different jobs

The Cambridge Business English Dictionary defines dictation as the activity of speaking something aloud for a person or machine to record ²⁹. That definition covers two jobs. Live dictation types as you speak. Transcription converts a recording into text afterward. Voice control is a third job: it operates the computer. The table sets them apart.

Dictation and voice control overlap on a Mac. In macOS 27, Apple says that when Voice Control is turned on, you use it to dictate text, and standard macOS Dictation is not available ³⁰. Inside Voice Control, Dictation mode is on by default ³¹. Command mode works differently: Voice Control responds only to commands, and words that are not commands are ignored rather than typed ³². Our dictation versus voice control versus Siri comparison explains which to use.

Commands can also be faster than the mouse for some work. In a 1992 study of word-processing tasks, speech-activated commands were significantly faster, with an average reduction of 18.67 percent in task time ³³. Participants made more memorization errors with speech than with the mouse ³⁴.

Dictation is speech to text, so its reverse is text to speech, which reads text aloud. Some dictation products include both. Dragon's Read That feature, for example, lets Dragon read editable text aloud in a computer voice ³⁵. The question on speech to text and text to speech sets out the difference. Transcription is a separate buying decision, covered in our guide to transcription software.

Modern dictation runs neural speech recognition, then optional AI cleanup

The first recognizers handled very small vocabularies. In 1952, Bell Laboratories built a system that recognized isolated spoken digits from a single speaker ³⁶. IBM's experimental Shoebox, designed in 1961, recognized 16 spoken words ³⁷. Today's systems use neural networks trained on large amounts of audio. OpenAI's Whisper models, for example, were trained on 680,000 hours of multilingual and multitask audio with transcripts ³⁸. Our guide to how speech recognition works goes deeper on the models.

Google said in 2019 that its new on-device recognizer design offered the same accuracy as its traditional server-based models ³⁹. That design is what lets recognition run on a phone without a connection.

Every dictation product follows the same path, with different parts running on your device or on a server:

Punctuation is a separate step. By default, Google Cloud Speech-to-Text leaves punctuation marks out of recognition results ⁴⁰, so developers must ask for it. Apple's Dictation on a Mac adds certain punctuation automatically in supported languages ⁴¹.

The cleanup step is where AI can enter. In the HyPoradise benchmark, a large language model could correct words missing from the recognizer's candidate list, because it generates text rather than choosing only among candidates ⁴². Apps sold as "AI dictation" can add a second model that rewrites or formats the raw words. That adds a second place where text can change, so check the output.

Accuracy is measured as word error rate (WER). In a 2025 pilot in neurosurgery, a modified Whisper-based model had a mean word error rate of 1.75 percent, counting all errors ⁴³. Dragon Medical One had 1.54 percent on the same measure ⁴⁴, and the difference was not statistically significant ⁴⁵. The pilot used three physicians ⁴⁶, so read it as a small study.

The microphone matters too. In a 2017 study with real recordings, a single distant ceiling microphone gave a word error rate roughly 3.2 times the close-talking rate, though the two conditions were not identical ⁴⁷. A 1996 study found that recognition performance degrades rapidly in ambient noise ⁴⁸. A close headset is the simplest fix.

Accuracy also varies between speakers. Across five commercial systems in a 2020 study, the average word error rate was 0.35 for Black speakers and 0.19 for white speakers ⁴⁹. The gap is a reason to test a tool with your own voice, not to trust a headline figure.

Speech recognition is not the same as speaker recognition. The second identifies who is speaking. The IEEE Technology Navigator describes speaker recognition as identifying or verifying the identity of an individual from characteristics of their voice ⁵⁰. The question on voice recognition and speaker recognition explains why the two get confused.

People use it to write faster, to avoid typing pain and for access

The clearest speed evidence comes from a controlled study on an iPhone. English text entered by speech reached 153 words per minute, against 52 words per minute on the built-in keyboard, 2.93 times faster ⁵¹. The chart compares the two methods.

That study used short phrases, read rather than composed, in a quiet room, with a 2016 recognizer ⁵¹. The lab figure leaves out editing, so real writing is likely slower. Correction can take most of the time. In a preliminary 2001 study with a small sample, users spent over 65 percent of their dictation time correcting errors and less than 35 percent dictating ⁵².

Skill also builds with practice. In a 1978 IBM Research study of dictation to a recorder, not to recognition software, novice dictators learned within a few hours to dictate letters as fast and as well as they wrote them, although they did not believe they were doing that well ⁵³. In a longitudinal study, users became more efficient at dictation, navigation and error correction as they gained experience ⁵⁴. The major improvements came from better dictation quality and from navigation commands ⁵⁵.

A dictated draft is not a finished document. In a 2018 study of clinical notes, 96.3 percent of the speech recognition drafts contained at least one error ⁵⁶. Their error rate was 7.4 errors per 100 words ⁵⁷. Review and editing cut that to 0.3 errors per 100 words, about 25 times fewer ⁵⁸. Dictate first, then read and correct.

For people who find typing painful or hard, dictation reduces how much they type. Microsoft's Voice Access is the successor to Windows speech recognition. Microsoft says Windows speech recognition is no longer being developed and is being replaced by Voice Access ⁵⁹. Its first-time setup needs an internet connection ⁶⁰. Our accessibility guides cover how dictation fits different needs, starting with dictation for RSI and carpal tunnel. The guides describe what the tools do. They do not give medical advice.

The kind you need follows from where you write and what you dictate

Start with the free tool in the app where you write most. Pay for a second tool only for a gap you have actually found. Five checks settle most decisions.

1. Confirm your language is supported. Apple lists 63 language and regional variants for Dictation on the Mac ⁶¹. Check your language and region before you rely on a tool.
2. Test the microphone. Use a close headset in the room where you will dictate. The distant-microphone figures above show why ⁴⁷.
3. Add your own terms. Names, clinical terms and product names are where recognition fails most often. Look for a custom vocabulary or word list before you buy.
4. Find out where the audio goes. On a Mac, the keyboard settings show whether dictation is processed on the device ⁶². Apple says dictation sent to its servers is not stored unless you opt in to Improve Siri and Dictation ⁶³. For cloud apps, read the policy and any training setting. Wispr Flow says turning off its model-improvement setting means nothing you say is used to train its models ⁶⁴.
5. Test on your own writing. Dictate a page of real work in your real apps, then count the corrections. Your result beats any ranking.

Yap, which we make, is an on-device dictation app for Mac and Windows, and it is one option among several. For many people the free built-in tools are enough, and our guide to choosing a dictation app explains how to compare them on your own writing. For the individual products, read our best dictation software guide.

Before you put dictated text into a document that others will rely on, proofread it. For the method, read the question on how to proofread dictated text.