Is there a speech to text API for Windows apps?

Yes. Windows apps can call several speech recognition APIs, but they do not all work offline. The UWP speech API uses predefined dictation grammars that run online, so an offline app has to use a grammar installed on the device instead. For full on-device speech to text, Microsoft's Embedded Speech SDK is designed for that job, but it needs approval, and heise reported in June 2026 that a newer local speech API was in public preview. Check each API's current status before you commit to it.

The UWP speech API runs dictation online

The UWP speech recognition API, which apps use through SpeechRecognizer, offers predefined dictation grammars. Microsoft says these grammars are online and not on the device, so they can be slower than a custom grammar ¹. A custom grammar is one you write for your app, with the phrases it should expect. That is the offline route inside this API, and it suits fixed commands better than free dictation.

If your app must dictate free text with no network, the predefined dictation grammar in this API is not the right choice. Test it with the network off before you rely on it.

Embedded Speech is built for the device, but access is limited

Microsoft's Embedded Speech is designed for on-device speech to text and text to speech, where cloud connectivity is intermittent or unavailable ². It requires Windows 11 or newer on x64 or Arm64 hardware ³. Microsoft limits access to the feature, and developers can apply for access ⁴.

For Python, the embedded package is named azure-cognitiveservices-speech-embedded, and it includes the cloud scenarios as well as the embedded ones, according to Microsoft's embedded speech page. Install only one of the two Speech SDK packages, as that page says.

Azure Speech containers can run offline only after approval

Speech containers are an alternative for environments that must stay disconnected. You must submit a request form and wait for approval before you can run them disconnected from the internet ⁵. "Offline" here does not mean zero contact with Azure: the containers still send billing information to Azure through a Foundry resource for Speech on your account ⁶. Read Microsoft's terms before you design an offline product around them.

A new Windows speech API runs on the device, but English came first

In June 2026, heise reported that Microsoft's new Windows speech recognition API converts speech to text on the device and does not need an internet connection ⁷. The public preview supported only English at first ⁸. heise reported Microsoft's Build 2026 claims, so treat these as a secondary report. We could not confirm the API's SDK name or namespace from Microsoft's own pages, so look for it on Microsoft Learn before you build on it.

What we did not find

We did not find a Microsoft statement on whether the desktop .NET engine, the SpeechRecognitionEngine class in System.Speech, works without an internet connection. Test it with Wi-Fi off, using the language packs installed on your PC, before you choose it for an offline app.

We also found no published word error rate for any of these Windows APIs, so there is no accuracy figure to compare them by. Measure accuracy on your own speech. For what the score means, see word error rate.

Choose by what your app must do

What to do

1. Decide whether the app must work with no network at all. If it must, the predefined dictation grammar in the UWP API is out.
2. For commands, write a custom grammar and test it with Wi-Fi off.
3. For free dictation, compare the Embedded Speech SDK, which needs approval, with the new local API, which is in preview and English first.
4. Check the vendor's current documentation on the day you build, because these APIs are changing.

For the terms behind these choices, see the Windows Speech API (SAPI), on-device speech recognition and Azure AI Speech. If you are looking for a service rather than building an app, see free speech to text APIs and whether speech to text APIs store your audio. For end-user Windows dictation, see our guide to offline speech to text on Windows.