Read this sentence and notice the voice saying it. That is subvocalization: the internal speech, and the faint muscular activity in the larynx and tongue that accompanies it, which most people produce while reading silently.
It is measurable. Attach sensitive electrodes to the speech muscles of a silent reader and you get signal, rising when the text gets harder.
Speed-reading instruction has treated this as the enemy for a century, on a simple argument: you can read faster than you can talk, so the voice is a speed limit. The argument is half right, and the half it gets wrong matters.
What subvocalization is actually doing
It is not a bad habit left over from learning to read aloud. It is how written words get into the part of working memory that holds language.
In the standard model of working memory, verbal material is maintained by a phonological loop, essentially a rehearsal mechanism operating on sound. Text arrives visually and gets recoded into that phonological form. Subvocalization is that recoding happening.
Which predicts exactly what is observed: subvocalization increases when text is difficult, unfamiliar, or syntactically complex. It is not a tic that harder text fails to suppress. It is a resource being recruited because it is needed.
The honest cost-benefit
| With subvocalization | Suppressed | |
|---|---|---|
| Reading rate | Roughly capped near speech rate, about 250–300 wpm for most | Can go faster |
| Comprehension of easy text | Fine | Often fine |
| Comprehension of hard text | Fine | Drops, sometimes sharply |
| Holding a long sentence together | Supported | Impaired |
| Retention | Better | Worse |
The pattern is consistent: suppression helps most where it matters least. On simple, familiar, low-stakes text you can skim without the voice and lose little, because you were not doing much processing anyway. On dense text, the voice is the machinery.
This is the same conclusion speed reading reaches from the other direction: past roughly 500 words a minute, what is being sold as reading is skimming, and comprehension measurements collapse.
What to do instead of eliminating it
The useful goal is not a quieter voice, it is matching your reading mode to the text.
Skim deliberately, and know you are skimming. For finding out whether a document deserves attention, for locating a section, for a familiar news article, go fast and accept shallow processing. Nothing is lost because nothing was going to be retained.
Slow down where it counts. For an argument you need to evaluate, a contract, a paper, or anything you will be tested on, reading speed is not the constraint. Understanding is. Trying to accelerate here trades the thing you want for the thing you can measure.
Read hard passages aloud. Counterintuitive if you are trying to be fast, but it is the strongest version of the same mechanism, and it also recruits the production effect. This is why read-recite-review works.
Reduce rereading instead. Regressions, the involuntary jumps back to earlier words, cost more time than the inner voice does for most people, and they usually signal that comprehension has already broken down. Slowing slightly often reduces them and nets out faster.
Where the speed-reading advice came from
The techniques are older than the evidence. Eliminating subvocalization, widening the perceptual span to take in whole lines, and using a finger as a pacer were all developed before much was known about how reading works, and they were marketed hard.
Of the three, the pacer survives scrutiny best, and mostly because it reduces regressions rather than because it raises raw speed. The claim that you can read several thousand words a minute with full comprehension does not survive testing at all.
The honest summary: reading speed is largely determined by how quickly you can understand, and understanding is not accelerable by technique. What is improvable is deciding what to read, and what to read carefully.
The note-taking connection
If comprehension is the bottleneck rather than speed, then the leverage is not in reading faster, it is in what you do with the text.
Reading a chapter twice at speed is worse than reading it once and recalling it. That is the single most robust finding available on studying, and it makes the whole subvocalization argument somewhat beside the point. See active recall, the testing effect, and how to remember what you read.
The one place suppression genuinely helps is a case few people think about: because the phonological loop is finite, holding a sentence in your head and writing a note at the same time compete for the same resource. That is part of why transcribing during a lecture damages comprehension, and why how to take lecture notes argues for fragments and questions rather than sentences.
More in working memory, cognitive load theory, and memory and focus.