Does Windows Voice Access use AI?

Windows Voice Access recognizes speech on your PC. Microsoft describes its recognizer as "modern, on-device speech recognition" ¹, and the overview of Voice Access on Microsoft's support site does not use the word AI. So the answer depends on what you mean. If AI means a learned model that turns sound into text, that describes many modern recognizers, but Microsoft does not say which kind Voice Access uses. If it means a generative assistant that writes or reasons for you, Microsoft's Voice Access pages describe no such feature.

The AI that Microsoft does name is fluid dictation on Copilot+ PCs, which uses a small language model to correct your text ². This is fluid dictation, not the recognizer. That feature is covered in our answer on what fluid dictation is. Voice Access itself is covered in our guide to Windows 11 Voice Access.

Voice Access recognizes speech with an on-device model

Microsoft's overview says Voice access "uses modern, on-device speech recognition to accurately recognize speech and works even without the internet" ¹. Two parts of that sentence matter. "On-device" means the recognition runs on your PC rather than on a server. "Modern" is Microsoft's own word, and the page does not say what kind of model the recognizer uses.

At first setup, Voice Access downloads the language files for your chosen language, and it needs an internet connection to do so ³. After that one download, the speech model runs on your PC, as Microsoft's on-device wording says.

Microsoft does not describe a generative model in these pages. Voice Access lets you control apps and write text by speaking, and it does not say that it writes sentences for you.

Many modern speech recognizers are neural networks, but Microsoft does not say so for Voice Access

Many modern speech recognizers are learned models. Google described its 2019 on-device recognizer, used for voice input in its Gboard keyboard, as an "end-to-end, all-neural, on-device speech recognizer" ⁴. "All-neural" means the recognizer is built from neural networks, a kind of machine learning. The Voice Access pages we read do not say which architecture its model uses, so we cannot say whether it is one. The on-device claim is the part Microsoft states clearly.

Even so, the word AI is used for a narrower set of features. Here is how Microsoft's own pages describe each Windows speech tool.

The table shows that Microsoft reserves the language of language models for fluid dictation. Voice typing uses a cloud service, and Voice Access uses a local recognizer that it does not label as AI.

Fluid dictation is where Microsoft uses language-model terms

Microsoft's voice typing help names the AI Phi Silica model as the download you can watch before fluid dictation is ready ⁵. The download is for Phi Silica, the AI model Microsoft names on that page. Fluid dictation runs on Copilot+ PCs. Microsoft says many of its new Windows AI features require an NPU that can run at 40 or more trillion operations per second ⁶. You can read more in our guide to Copilot+ PCs and Windows on Arm.

This is the part of the Windows speech stack that clearly uses language models. A PC without the Copilot+ label does not get fluid dictation, so it does not get that model either. Voice Access itself is available on Windows 11 version 22H2 or later ⁷, and Microsoft does not tie it to the Copilot+ label.

Accuracy is a separate question from AI

Calling a feature AI says nothing about how accurate it is. Microsoft says Voice access "might make mistakes in recognizing the text dictated by users" ⁸. If a name or technical word keeps coming out wrong, Voice Access has an Add to vocabulary command for improving recognition of complex or hard to pronounce words ⁹.

Accuracy also depends on your microphone, your room and how clearly you speak. Our guide to Windows dictation problems covers the usual causes, and the question on Voice Access and Windows Speech Recognition explains how the older tool differs.

Privacy depends on which tool you use

Whether AI is involved matters less for privacy than where the audio goes. Voice Access recognition is described as on-device, and it works without internet after setup ¹. Fluid dictation's cleanup is also on the device ¹⁰. Voice typing's recognition is different: it is an online service ¹¹.

At its Build 2026 developer conference, Microsoft unveiled a new speech recognition API, according to a June 2026 report from heise ¹². The report says that "Processing runs directly on the device and does not require an internet connection," but it does not say that Voice Access uses that API. The same report says the public preview "initially only supports the English language" ¹³. Our guide to Windows dictation privacy explains how to choose between these tools.

What to do

1. Decide what you mean by AI. If you want to know whether speech is processed on the PC, Voice Access is described as on-device after its one-time language download.
2. Turn on Voice Access from Settings or by searching for it in Windows search, and let it download your language before you go offline ¹⁴.
3. Test it on your own speech for a few minutes. Expect some errors, and correct one word at a time ¹⁵.
4. If you want the language-model cleanup, check for the Copilot+ label first, since fluid dictation does not run without it.

The short answer for most readers is this: Voice Access recognizes speech on your PC, Microsoft does not say whether its recognizer is machine learning, and the language-model AI that Microsoft names belongs to fluid dictation.