Can you train Voice Access to recognize your voice?

Yes, but only in a limited way. Voice Access has no general training session where you read a set passage so it learns how you speak. Its one voice-learning step is Voice Isolation, a quick one-time setup that Microsoft says lets Voice access learn to recognize your voice ¹. It creates a voice signature on your PC ², and Microsoft describes its job as filtering out other speakers and background noise. You can re-record that signature or delete it. For words the feature keeps getting wrong, you add them to the vocabulary.

Setup is optional. Voice Access works without it, and it can make mistakes either way. Whether the setup is worth the time depends on the room you dictate in and on how you speak.

Voice Access has no general training session

We found no session in Microsoft's Voice Access pages that trains the recognizer on your everyday speech. The feature starts working once the language setup finishes, and you can use it straight away. The only voice-specific step is the Voice Isolation setup described below.

Older tools are different. Windows Speech Recognition, the feature Voice Access replaced, is a separate program, and this page does not cover its settings. For that tool, see our guide to Windows Speech Recognition.

Voice Isolation is the one step that learns your voice

To set up Voice Isolation, open the Voice access settings menu on the Voice access bar, go to Improve speech recognition, and select Voice Isolation. The setup then has three parts:

1. Download the voice model. Setup needs internet for this download ³. The model turns your recording into a voice signature.
2. Read the paragraph. Select your preferred microphone and begin reading the paragraph aloud ⁴. Microsoft's tips say to sit in a quiet place, keep the microphone at a comfortable distance and speak at your normal pace.
3. Finish. When the recording is done, you see a success message confirming that Voice Isolation has been enabled.

Microsoft says the model runs on your device and that no raw voice recording is saved. Only the voice signature is stored locally ².

The point of the setup is filtering, not accuracy in general. Voice Isolation can still pick up someone who speaks at the same time as you ⁵. Microsoft does not publish a figure for how much the setup improves recognition, and we found no independent test of it. Treat it as a help in shared rooms, not as a guarantee.

You can re-record or delete the voice signature

After setup, the Manage Voice Isolation menu lets you change what the feature knows about you. If Voice Isolation does not perform as well as you expect, Microsoft says you can record a new voice sample to improve the experience ⁶. That is the first thing to try if the first recording was made in a noisy room.

You can also delete the recording. Microsoft says this removes your voice signature from the device ⁷.

Vocabulary teaches names and jargon, not your voice

Voice Access also has an Add to vocabulary option. Microsoft lists it for improving the recognition of complex or hard to pronounce words ⁸. This is a different kind of help. It does not change how the feature hears your voice. It teaches the recognizer a word you use often, such as a name, a product or a technical term.

Use vocabulary for individual words. A short list of the words that keep coming out wrong is more useful than a long one, and it is quicker to check.

It depends on the room, your speech and your text

The right choice depends on three things. The first is where you dictate. The Voice Isolation setup is worth the time in a shared office or a noisy room. In a quiet room, you may not need it at all.

The second is how you speak. We found no Microsoft statement on how Voice Access performs for people with speech that differs from typical speech, so if that applies to you, test the feature before you rely on it.

The third is the text you need. Voice Access shows its transcription as you speak, so read it as you go, especially on long passages.

What to do

1. Test first. Read a paragraph you know well into Voice Access, then check the transcription on the bar. If the errors are few, you may not need Voice Isolation at all.
2. If the same words keep coming out wrong, add them to the vocabulary, one word at a time.
3. If the room is noisy or other people speak nearby, turn on Voice Isolation from the Improve speech recognition menu, and pick the filtering option that fits the room.
4. If the result is still poor, re-record your voice sample once in a quiet place. If that does not help, delete the recording and use the other filtering options.

For the first steps of setup, see our guide to Voice Access. If you want the basics of what the feature does, read what Voice Access is. For the terms used here, see the glossary entries for voice profile and custom vocabulary.