Dictation: Unprompted word repeatedly appearing.

Regardless of the recording software or microphone I use, spurious words appear while I'm dictating, even when I'm paused.
For example, if I say, "Hello, I have a question for you", what might appear is "Hello, will I have a will question for you will". It is always "will". (The italics are to help highlight the spurious word.)
Anyone have a clue what is causing this?

Winver: 22H2, build 22621.6060

My first question is why are you still running unsupported Window 10 22h2. This IS a Windows 11 forum.

Per co-pilot here are the causes and the fixes.

The behavior you’re seeing — the word “will” being inserted repeatedly even when you’re paused — points to a corrupted Windows speech profile or a stuck dictation language model. Your example shows the recognizer injecting “will” between phrases and even at the end of the sentence , which is exactly what happens when the speech engine falls back to a single high‑confidence token.

Since it happens regardless of the recording software or microphone , the issue isn’t hardware or app‑specific — it’s Windows’ speech subsystem.

Fixes that actually resolve this:​

1. Reset the Windows Speech Recognition profile Control Panel → Speech Recognition → Advanced → Delete Speech Recognition ProfileThis clears corrupted acoustic/language data.

2. Reset Online Speech Recognition Settings → Time & Language → Speech → turn Online Speech Recognition off → reboot → turn it back on.

3. Remove extra English language packs If you have multiple English variants installed, remove all but one.This prevents token‑mapping conflicts.

4. Restart stuck dictation processes End task:

• ctfmon.exe
• SpeechRuntime.exe
• TextInputHost.exe Then restart dictation.

5. Test with Windows Speech Recognition (WSR) If WSR also outputs “will” repeatedly, the speech profile is definitely corrupted and step 1 fixes it.

Why this happens​

Windows 10’s dictation engine (you’re on 22H2 build 22621.6060) can get into a state where a single token is emitted whenever the recognizer is idle or uncertain. “Will” is a common fallback token in the English model.

Resetting the speech profile and dictation engine clears the stuck token.