This is a bit ridiculous.

I can count out loud while reading (doing both at maximum speed for each separately). This means that the motor areas in the brain which generate signals to control muscles in my tongue, larynx, etc. compute the right nerve impulses for words 1, 2, 3, 4, 5 .... . This means that they are not computing the nerve impulses for vocalization of what ever I am reading.

Now, if during reading some little signals are present, that can mean all sorts of things. E.g. in a robot that would mean that the power supply's voltage stabilization is not perfect. Unless they actually decode words from the impulses they can't say if its some sort of noise or subvocalization.

That claim does not seem to appear in the two sources he cites. The word subvocalize does not appear in Rayner and appears only twice in Carver. Going by the titles, it would be rather surprising if it appeared in those sources. I'm skeptical enough of the usefulness of the claim that I'm not going to try other sources, but if someone gives me a precise citation, I'd be interested.

I think private messaging makes an interesting point, but I'm more concerned with hearing (or noticing that your hear) the voice in your head saying the words.

In other words, even if you are making incredibly small movements with your lips/throat, can you reduce only the subjectively observable mental voice, and how can it affect your reading speed or experience.