Speech to text guide: how it works and how to start

A practical speech to text guide: what it is, the free tools on Mac, Windows, iPhone and Android, cloud vs on-device, what affects accuracy, and how to pick one.

By The Sayburst team · · 6 min read

Speech to text turns what you say into written text, and you probably already have it. Every Mac, Windows PC, iPhone and Android phone ships with a free version: on a Mac it's Dictation in System Settings › Keyboard, on Windows it's WindowsH, and on a phone it's the microphone key on the keyboard. Start with that, then move to a dedicated app if you dictate every day and want cleaner text or offline privacy.

This guide covers what's actually going on, which built-in tools exist where, the choices that matter, and how to get good results in your first week.

Dictation and transcription are different jobs

Both are speech to text, but you use them differently.

  • Dictation is live. You talk, and text appears where your cursor is: in an email, a chat, a document. It replaces typing.
  • Transcription works from a recording. You hand over an interview, a lecture or a voice memo, and get a transcript back. It replaces listening and typing it up.

Most of the confusion about "the best speech to text" comes from mixing the two. Our article on dictation vs transcription goes deeper.

The free speech to text you already have

Device or appHow to startWhere recognition happens
MacSystem Settings › Keyboard › Dictation, then your shortcut or Edit › Start DictationOn the device or Apple's servers; the Dictation settings show which
Windows 10 and 11Windows key + H in any text boxOnline (Microsoft's Azure Speech services)
iPhone and iPadMicrophone key on the keyboardSee Apple's iPhone User Guide
Android (Gboard)Microphone at the top of the keyboardNot stated in Google's Gboard help
ChromebookTurn on Dictation in Accessibility settings, then Search + DNot specified by Google
Google DocsTools › Voice typingGoogle's online service
Microsoft WordHome › Dictate (Microsoft 365)Online; needs a connection

From Apple, Microsoft and Google support pages, checked October 31, 2026. Apple's Mac Dictation guide, Microsoft's Windows voice typing page and Google's Gboard voice typing help have the details for each.

The system-level tools (Mac Dictation, Windows voice typing, phone keyboards) work in every app. The app-level ones (Google Docs, Word) only work inside that app.

Cloud or on-device: the choice that matters most

Speech to text runs either on a server or on your own computer.

Cloud recognition sends your audio to the provider, which returns text. It needs a connection, and your voice leaves your device. Many big services work this way.

On-device recognition runs the speech model on your own hardware. Nothing is uploaded, it works offline, and there's no per-minute cost. Apple silicon Macs are fast enough that this is now practical for everyday dictation.

Neither is wrong. If you dictate client details, health information or anything you'd rather keep to yourself, on-device is the easier answer. Our local vs cloud dictation guide breaks down each step where data can travel.

What actually affects accuracy

In rough order of impact:

  1. The microphone. A headset near your mouth beats a laptop mic across the room every time.
  2. Background noise. Other voices are the worst kind, since the software can't tell they're not you.
  3. How you speak. Whole sentences at a normal pace work better than slow, word-by-word dictation.
  4. Vocabulary. Names, brands and jargon are where general models slip. Tools with custom vocabulary fix most of this.
  5. The model. Modern models are close on clean speech; differences show up in noise, accents and unusual words.

On models: two families you'll see a lot on the Mac are OpenAI's Whisper and NVIDIA's Parakeet. In our own benchmark on an M4 Max, on a 1,082-word sample of clean read speech from LibriSpeech, Parakeet v3 had a 0.46% word error rate with a median of 82 ms per dictation, and Whisper Large v3 Turbo 0.83% at 320 ms. That's clean read speech, not a noisy office, so treat it as a comparison between models rather than a promise. The full method is in Parakeet vs Whisper on Apple silicon.

Getting started on a Mac

  1. Turn on Dictation in System Settings › Keyboard and set a shortcut you'll remember.
  2. Choose your microphone under Microphone source.
  3. Click into TextEdit and say a few sentences. Apple adds punctuation automatically in supported languages.
  4. Try it where you really write: a reply in Mail, a message in Slack.
  5. Give it a week before deciding. Our guide to how to dictate on a Mac covers the details, and using speech to text in your daily workflow shows where to start.

Common problems

  • Nothing happens: dictation isn't on, the shortcut conflicts with something, or the wrong microphone is selected.
  • Text is in the wrong language: check the dictation language setting and switch it to the language you're speaking.
  • It stops on its own: Apple Dictation stops after 30 seconds of silence. Hold-to-talk tools avoid that.
  • Names are wrong: add them to custom vocabulary, or type them.
  • Text full of "um": built-in tools type what you say. Dedicated apps clean it up.

When the built-in tools are enough

If you dictate occasionally, the free tools are good. Mac Dictation in particular works in every app and costs nothing. For phones, Windows and Chromebooks, the built-in options are the right place to stay; Sayburst is a Mac app and doesn't run there.

When to use a dedicated app: Sayburst on the Mac

If you write by voice every day on a Mac, Sayburst is built for that. Hold fn, talk, let go: speech is transcribed on your Mac with Parakeet or Whisper, cleaned up, and inserted wherever your cursor is. Audio is never uploaded, and after the model downloads once, it works with no internet.

A text to a friend, in Message mode:

"running about ten minutes late, um, the train's stuck outside, outside Clapham. Order me whatever you're having"

What gets sent:

Running about 10 minutes late, the train's stuck outside Clapham. Order me whatever you're having

Fillers and the repeated word are gone, and Message mode keeps it casual, without a formal final period. In Mail, the same tool can produce a properly laid-out email; in Slack, a quick reply. You can add your own vocabulary, choose between several Parakeet and Whisper models, and Whisper covers 99 languages. Comparing options? See the best talk to text apps.

Sayburst needs an Apple silicon Mac on macOS 14 or later. It's $29 once for one Mac after a 7-day free trial, with no subscription.

Try speech to text that stays on your Mac. Free for 7 days, no card.

Download Sayburst

The short version

  • Occasional use on any device: the built-in dictation.
  • Recordings and interviews: a transcription tool.
  • Daily writing on a Mac, private and offline, with clean text: Sayburst.

Hold fn, talk, and clean text lands where your cursor is. Free for 7 days.

Download Sayburst

Questions people ask

What is speech to text?

Software that turns spoken words into written text. You either talk and watch the text appear where you're typing, which is dictation, or give it a recording and get a transcript back, which is transcription.

What is the best free speech to text?

The one already on your device: Dictation on a Mac, voice typing with Windows key + H on Windows, the microphone key on iPhone and Android keyboards, and Voice typing in Google Docs. Start there before paying for anything.

Does speech to text need the internet?

It depends on the tool. Windows voice typing and Google Docs voice typing use online recognition. Some Mac and phone dictation runs on the device, and apps built on models like Whisper or Parakeet can run entirely offline.

How do I make speech to text more accurate?

Use a close microphone, reduce background noise, speak in whole sentences, and add names and jargon to the tool's custom vocabulary if it has one. The microphone usually matters more than the software.