Windows Speech Recognition: Deprecation and Replacement

If you use Windows 11 version 22H2 or later, Voice Access is the built-in replacement for Windows Speech Recognition. Microsoft says Voice Access replaced the old tool on Windows 11 version 22H2 and later in September 2024 ¹. If you use Windows 10, the old tool is still there, but Windows 10 reached the end of support on October 14, 2025 ², and it no longer gets security updates ³. Plan your move to Voice Access now, and treat the old tool as a short bridge.

This guide covers the dated timeline from Microsoft's own lists, what Windows Speech Recognition did that Voice Access does not, how to move a long-standing workflow across, and which speech APIs developers can still build on. For the current voice tools, start with our overview of dictation on Windows and the guide to Voice Access in Windows 11.

Windows Speech Recognition was Windows' built-in voice control tool, with a training wizard and a dictionary

Windows Speech Recognition, often called WSR, let you dictate text and control parts of the PC by voice. Its setup had two features that people relied on for years. The first was voice training. On Windows 10, Microsoft's help says: "You can teach Windows 10 to recognize your voice." ⁴ The help article walks you through a Set up Speech Recognition wizard for that.

The second was the Speech Dictionary. Microsoft's help for the tool says you can add words that are frequently misheard or not recognized ⁵. People who dictate names, drug names, product codes or jargon often depend on this list, and it is the first thing to rebuild when you move, as covered in the migration section below.

Some of Microsoft's own documentation for the old tool was written for Windows 7, so menu names on current versions may differ from what those pages show.

Microsoft deprecated it in December 2023 and named Voice Access as its replacement

Deprecation is the first step in retiring a feature. Microsoft's list of deprecated Windows features gives the month for the change: December 2023 ⁶. The entry says that Windows speech recognition "is deprecated and is no longer being developed", and that the feature "is being replaced with voice access" ⁷.

Deprecation does not mean the tool disappeared that day. It means Microsoft stopped adding to it. The same deprecated-features entry names the replacement and its minimum Windows version ⁸.

The first two rows come from Microsoft's deprecation list and its command reference. The third row is the Windows 10 date, which matters for anyone still on that system. The gap between the first two rows was nine months ⁹.

Voice Access took over on Windows 11 version 22H2 and later in September 2024

Microsoft's command reference for the old tool says it plainly: "Voice access replaced Windows Speech Recognition (WSR) for Windows 11 22H2 and later versions in September 2024." The same note adds that WSR is available in older versions of Windows ¹.

So the split is by Windows version, not by individual PC. A Windows 11 PC on 22H2 or later gets Voice Access. An older Windows 11 build keeps the old tool. Microsoft does not give an exact day or build number for the change on these pages, so if you need to know which update removed the tool on your PC, check the release notes for your own build.

Voice Access is available in Windows 11, version 22H2 and later, according to Microsoft's deprecation entry ⁸. That entry lists Windows 11 devices only, so Voice Access is not a Windows 10 option.

The two tools differ on training, words, languages and offline use

The table below sets the main differences side by side. Each row reflects what Microsoft's pages say, and where a page is silent we say so.

Voice Access needs an internet connection to complete its first-time setup ¹², including the download of its language files ¹³. After that, Microsoft describes Voice Access as a way to control your PC and write text "without an internet connection" ¹⁴. That is why it suits unreliable networks and speech you do not want sent to a server.

Windows 10 keeps the old tool, but Windows 10 is past end of support

Windows 10 still includes Windows Speech Recognition, and Microsoft's help still explains its setup steps. The operating system underneath has reached the end of support. Microsoft's page says that Windows 10 "has reached the end of support on October 14, 2025" ². Counting from that date to October 16, 2026, the day this guide was checked, Windows 10 had been out of support for 367 days ¹⁵.

A speech tool on a system that no longer gets security updates is a reason to plan a move to Windows 11, not a reason to stay. If you are staying on Windows 10 for now, follow the steps in our question on setting up speech recognition on Windows 10. If the tool will not start, see the question on why speech recognition is not working.

Voice Access runs on the device, and it is not the same tool as Windows voice typing

Voice Access has a different design from Windows voice typing. During setup, it downloads language files for the language you chose, and these files are used by on-device speech recognition ¹³.

Voice typing, the Windows key + H tool, is different. Microsoft says that voice typing "converts speech to text using online speech recognition" ¹⁶. Microsoft's privacy page also describes the device-based option in Windows 11: device-based speech recognition "Processes your voice locally on your device. No voice data is sent to Microsoft." ¹⁷

The two tools also conflict. If Voice Access is running and you press the Windows key + H, voice typing does not start, and Windows tells you to dictate with Voice Access instead ¹⁸. Close one before you use the other. For a fuller explanation of voice typing, see our guide to Windows voice typing.

Microsoft has not published an accuracy comparison

Many readers want to know whether Voice Access recognizes speech better than the old tool. We could not find an independent test, or a Microsoft benchmark, that compares the word error rates of the two. Some reviews describe Voice Access as the newer engine, but those are opinions, not measurements. If accuracy matters for your work, read the same short passage through both tools on the same microphone and count the errors yourself. That is the only fair test available today.

Voice Access covers more languages than the old tool, but Microsoft's pages disagree on the count

The language list is the clearest difference Microsoft documents. Windows Speech Recognition is "available only for the following languages: English (United States, United Kingdom, Canada, India, and Australia), French, German, Japanese, Mandarin (Chinese Simplified and Chinese Traditional), and Spanish" ¹⁰. Counting the names in that list gives 11 locales.

Voice access lists 15 languages and dialects: six English variants, two Spanish, German, two French, Simplified and Traditional Chinese, Japanese and Italian ¹¹. The difference between those two lists is four locales ¹⁹.

Microsoft's pages do not agree on Voice Access. The FAQ says voice access "is currently available in English, Spanish, French, and German" ²⁰, while the setup page lists 15 ¹¹. When the pages disagree, use the setup page and check the language menu on your own PC.

Language support is not accuracy. A longer list does not show that every language is recognized equally well, so test your own language before you rely on either tool.

Windows Speech Recognition's training and dictionary have no direct match in Voice Access

The two features people rely on most in the old tool are the ones that do not carry over in a simple way.

Voice training in Windows Speech Recognition was a setup step that taught the tool your voice ⁴. The Voice Access setup and help pages we read describe a speech model download ¹³, not a training session. Do not expect a voice profile from the old tool to move across.

The Speech Dictionary has a closer match. Voice Access calls it Add to vocabulary. Microsoft lists the command for improving recognition of complex or hard to pronounce words ²¹, which is the same purpose as the old list. The entries are separate, though, so plan to add your words again.

Moving to Voice Access means relearning a few commands and re-adding custom words

Moving a long-standing workflow is mostly a matter of re-creating three things: the switch, the words and the commands. Follow these steps on a Windows 11 PC on version 22H2 or later.

1. Open Voice access from Settings, or search for it in the Windows search box ²².
2. Let setup finish the download of your language. You need an internet connection for this step ¹².
3. Choose a mode. Voice Access works in three modes, chosen by the task you want to do ²³. In dictation mode you get no commands: every word you say is written into an editable field ²⁴, so use it when you want to write without triggering actions.
4. Re-add your custom words with Add to vocabulary ²¹. Start with the words you wrote down from the old Speech Dictionary.

Then relearn the correction commands from Microsoft's command list. Practice them on text you do not care about before you rely on them for real work. Voice Access also turns dictation into commands in some modes, so check the mode first when something unexpected happens.

Some people run a dedicated dictation app beside Windows' own tools when the built-in options do not fit their text. Yap, which we make, is an on-device dictation app for Mac and Windows, and you can read about it on the Windows page. Whichever tool you choose, test it with your own vocabulary first.

Developers still have Windows speech APIs on the device and Azure AI Speech in the cloud

If you built an application on Windows Speech Recognition or the classic speech API, you have several options. They differ in where recognition happens and in which Windows versions they suit.

SAPI is one of the native options. Microsoft's overview describes the software between applications and speech recognition engines, and it is written for an older generation of Windows, so check whether its guidance still matches the version you target. The .NET class is a practical choice for desktop applications written in .NET.

The UWP API has a limit to note. Its predefined dictation and web-search grammars are online, and Microsoft says performance might not be as fast as with a custom grammar installed on the device ²⁷. An application that must work offline should use a custom grammar.

For speech recognition that runs on the device through Azure's SDK, the option is embedded speech. On Windows, the requirement is Windows 11 or newer on x64 or Arm64 hardware ³⁰. Microsoft also limits access to it: "You can apply for access through the Azure Speech in Foundry Tools embedded speech limited access review" ³¹. Plan for that approval step before you commit to the design.

If you are unsure which option fits, our guide to speech to text APIs compares them across platforms, and our guide to on-device speech recognition explains what runs locally and what does not.

What to do next

Start by checking your Windows version before you change anything. If you have Windows 11 22H2 or later, turn on Voice Access and move your custom words while the old tool is still on the PC. If you have Windows 10, make a plan to move to Windows 11, because security updates have ended ³. For the built-in tools in more depth, read our guide to Windows voice typing, and if privacy matters for your work, read the guide to Windows dictation privacy.

For the terms used in this guide, see Windows Speech Recognition, Voice Access, online speech recognition, Windows speech pack and Windows Speech API.