Voice typing test

What this test measures

Two rounds, one passage of seventy-one words. Using the same text twice removes the two things that make most speed comparisons meaningless — different content and different length — so the only variable left is you.

Round 1: speaking words per minute

Press start, read the passage aloud at the pace you would use talking to a colleague, press stop on the last word. Because the passage length is known before you open your mouth, your speaking rate is simply seventy-one words divided by the time you took. The clock is the browser’s monotonic timer, so a system-clock change or a daylight-saving jump mid-round cannot distort it.

One deliberate choice worth knowing: speaking speed is calculated from the passage word count, not from the transcript. If the recogniser drops three words, that is a recognition failure and it is charged to the accuracy column — it does not quietly make you look slower than you spoke. The flip side is that the number assumes you read the whole thing, so if you give up half way the tool says so rather than reporting a flattering rate.

Round 2: typing words per minute

Type the same passage. The clock starts on your first keystroke, not when the page loads, so reading it through first costs you nothing, and it stops when you say you are done. Your speed is your own word count over your own elapsed time, which means racing ahead and leaving words out cannot buy a better number — the accuracy score catches it.

The headline typing figure counts real words. Conventional typing tests count five characters as one word instead, so that number is shown underneath as well; the two usually land within a few WPM of each other on ordinary English prose. If you want the keyboard number on its own terms, the typing speed percentile test and the keystrokes per hour test are built for exactly that.

How the accuracy score works

Both rounds are scored the same way. The passage and your attempt are reduced to plain words — lowercased, punctuation and apostrophes removed, hyphens treated as spaces — and then aligned with a longest-common-subsequence diff. That alignment matters: a position-by-position comparison would see one dropped word, shift everything after it, and report a catastrophic score for a near-perfect read. The diff keeps the two sequences in step, so a mistake costs one word.

Accuracy is the number of matched words divided by whichever sequence is longer. That denominator is what stops a transcript full of invented extra words from scoring well. Every mismatch is then printed back to you: words that never arrived are struck through, words that arrived but are not in the passage are underlined, and a swap shows both with an arrow between them. A bare percentage tells you nothing about what went wrong; “colleague” heard as “college” tells you a great deal.

Speaking words per minute test: the live meter

The second mode drops the passage entirely. Press start, talk about whatever you like, and watch your rate update as you speak — one big number for the last fifteen seconds, one for the session average, plus a running word count. It is the quickest way to find your natural conversational pace, and it is genuinely unnerving to watch it climb when you get enthusiastic about something.

Every word counts in that number, filler included. “Um”, “you know” and a restarted sentence inflate a speaking rate exactly as much as real words do, which is why a rate measured on free speech usually reads higher than the same person reading prose aloud. If what you actually need is the reverse calculation — a word count turned into minutes for a talk or a script — use the speech time calculator, which does words to minutes and minutes to a word budget at slow, natural and fast paces.

What counts as a good speaking WPM?

Rough guides, not verdicts. Unhurried presenting and audiobook narration tend to sit around 150 words per minute — slow enough to be followed by someone who has never heard the material. Ordinary conversation runs faster, commonly between 120 and 160. Reading a familiar passage aloud lands somewhere in the same band, usually a little under conversation because your eyes have to track the text. Above roughly 180 you are into the range where listeners start asking people to repeat themselves, and podcast listeners on 1.5× playback are already there.

Typing is the other half of the comparison, and the gap is the point. A competent office typist works somewhere around 40 to 60 words per minute on prose they are composing rather than copying; even a fast one rarely doubles that for long. Talking at 140 against typing at 45 is a little over three times the throughput, which is why the result panel converts your own two numbers into minutes saved per thousand words rather than leaving them as abstract rates.

Treat that saving as a ceiling. Dictating a real document also costs thinking time, and spoken drafts need a proofread — the honest saving sits below the projection and well above zero.

Which mode you get, and why

The transcription half of this test uses the Web Speech API, which today means Chrome, Edge, and other Chromium browsers such as Brave, Arc and Opera. Firefox does not implement it at all and Safari’s support is partial and prone to stopping after a few seconds. Rather than hand those visitors a dead button, the page detects what your browser can do and says which of the two modes you are in.

• Recognition mode — the microphone is requested when you press start, your read is transcribed, and you get speed and accuracy.
• Stopwatch mode — no microphone permission is requested at all. You press start, read aloud, press stop, and speaking WPM is calculated from the known word count. You get speed but no accuracy score, because nothing heard what you said.

Refusing the microphone permission, or having no microphone attached, drops the test into stopwatch mode automatically and tells you why. You can also switch to the stopwatch deliberately — useful in a noisy room, where the recogniser’s score measures the room rather than you.

One honest caveat about recognition mode: the accuracy you measure belongs to the browser’s speech engine, not to any particular desktop dictation app. In Chrome that engine streams your audio to Google’s servers. This page sends nothing anywhere — no upload, no account, no storage — but the browser does, so keep genuinely sensitive material out of any browser-based recogniser.

Getting a score that reflects you

• Read the passage once silently first. Both clocks start on a button or a keystroke, never on page load, so preparation is free. A cold-read stumble measures your sight-reading, not your speaking.
• Sit closer to the microphone than feels necessary. Laptop array mics pick up the room as readily as your voice, and half a metre of distance costs more accuracy than any accent does.
• Do not slow down for the recogniser. Over-enunciating actually hurts: these engines are trained on natural connected speech, and dictating one word at a time removes the context they use to choose between similar-sounding options.
• Run it twice. First runs are self-conscious. The second is closer to how you talk.
• Compare like with like. If you type the passage while looking at it, read it aloud while looking at it too. Copying and composing are different tasks at different speeds.

Where VoiceSnap Pro fits

This page exists because a dictation speed test is the fastest honest argument for dictation. VoiceSnap Pro is a voice-to-text app for macOS and Windows: you hold one keyboard shortcut, speak, and clean punctuated text appears in whatever field your cursor is already in — the email, the doc, the ticket, the pull request. It adds punctuation and paragraph breaks, strips “um” and “uh” as it goes, handles 50+ languages, and saves every dictation to a searchable notes library. It is a one-time $39 purchase rather than a subscription.

It has not shipped yet. Join the waitlist for one email on release day. In the meantime the browser speech-to-text tool is the same Web Speech engine used here, wired up for actual dictation rather than a test.

Questions people ask

Is my voice recorded or uploaded by this page?

The page records nothing and stores nothing: no audio file is created, no transcript is saved, and closing the tab discards everything. In recognition mode the browser itself handles the audio, and Chrome’s implementation sends it to Google’s speech service to be transcribed. In stopwatch mode the microphone is never touched at all.

Why is my accuracy score lower than I expected?

Usually the microphone rather than the speech. Distance, a fan, a hard-surfaced room and a cheap built-in array all cost more than accent does. Check the diff underneath the score: scattered single-word swaps point at audio quality, while whole missing runs usually mean the recogniser stopped and restarted mid-sentence.

Is a browser speech test representative of a real dictation app?

Partly. The speed half is entirely representative — how fast you talk does not change with the software. The accuracy half is a measure of this browser engine on this microphone in this room, and desktop apps that run a dedicated model, apply their own punctuation and let you add custom vocabulary for names and jargon are working with advantages the browser API does not have.

Why does my typing WPM here differ from other typing tests?

Most typing tests count five characters as one word and penalise errors as you go. This one counts real words for comparability with speech, scores accuracy separately after the fact, and shows the five-character figure underneath so you can line it up with the test you are used to. Passage length matters too: a seventy-one word sample rewards a good start more than a three-minute test does.

Can I take the test in another language?

The passage test is English, because the passage is English and the accuracy score compares against it word for word. The live meter takes a language selection and works in any of the languages the browser’s recogniser supports — word counting is language-agnostic. Languages written without spaces, such as Chinese and Japanese, are the exception: a words-per-minute figure does not really apply.

Does the test work on a phone?

The stopwatch half works anywhere. Recognition works in Chrome on Android; on iOS every browser uses Safari’s engine underneath, so the passage test will usually run in stopwatch mode there. Typing the passage on a phone keyboard is also not a fair test of your typing — that round is worth doing on a real keyboard.

How many words should the passage be?

Seventy-one, here. Long enough that a fumbled first second does not dominate the result, short enough to read twice without it becoming a chore. If you want to measure speaking rate over something longer and more natural, use the live meter and talk for two or three minutes instead — that number is closer to what you will actually sustain while dictating real work. To go the other way and turn a word count into minutes, the speech time calculator and the word counter both do the arithmetic.