Press Start listening, speak, and the words appear in the transcript box. The page uses the speech recognition built into your browser, so there is no account, no upload to ToolsHub4u and no recording kept after you close the tab.
The transcript is a draft, not a finished document. Recognition is good at ordinary sentences and weaker on names, numbers and specialist terms, and it punctuates by guessing where a sentence ended. The counters below the box help you judge the result: word count, sentence count, how long it takes to say and to read, and how many filler words survived.
Ten languages
The language list is fixed on the page, not read from your browser. It covers ten locales:
| Language | Code | Language | Code |
|---|---|---|---|
| English (United States) | en-US | Portuguese (Brazil) | pt-BR |
| English (United Kingdom) | en-GB | Italian (Italy) | it-IT |
| English (Australia) | en-AU | Hindi (India) | hi-IN |
| Spanish (Spain) | es-ES | Japanese (Japan) | ja-JP |
| French (France) | fr-FR | German (Germany) | de-DE |
Pick the language you will speak, not the language of the page. An English recogniser listening to Spanish produces confident nonsense. If the browser's engine does not support a locale, the panel reports that the language is not supported and stops; try another one or paste the transcript instead.
How a dictation session works
Recognition runs continuously with interim results. That means the panel shows a phrase in grey while you are still saying it, then commits it to the transcript when the engine decides the phrase is final. Interim text is not added to the box until it is final, so a phrase that changes on screen is normal rather than a fault.
Final phrases are appended to the transcript with a space between them. The Keep the transcript when starting again option decides what happens at the start of the next session:
- Ticked (the default). The existing text is trimmed and kept, and new speech is appended. Use this to dictate in several sittings.
- Unticked. The box is cleared the moment you press Start listening, so each session starts empty.
Press Stop when you have finished. The phrase in progress may still arrive if the engine returns it as a final result, so wait a moment before editing. Press Start listening again to continue; recognition does not resume on its own.
Sessions also end by themselves. Chrome and Edge stop after a stretch of silence, after a network problem, or when the tab has been in the background for a while. The panel says "Stopped" and the transcript is kept, so press Start again rather than assuming the page has broken.
The error messages
| Message | What happened | What to do |
|---|---|---|
| Microphone access was blocked | The browser denied permission | Allow the microphone in site settings, then reload |
| No microphone was found | None connected, or another app is using it | Connect one and close the other app |
| The recognition service could not be reached | Network failure | Check the connection and try again |
| Nothing was heard | Silence, or the microphone is too far away | Move closer and speak again |
| The browser refused the recognition service | Permission or policy block | Check site permissions, or try another browser |
| That language is not supported | The engine lacks that locale | Choose another language or paste the text |
The first session always asks for microphone permission, and that prompt comes from the browser rather than this page. If the Start button is disabled instead of showing an error, the browser does not expose the recognition API at all — Firefox is the usual case, and in-app browsers often behave the same way.
What the counters mean
Six statistics sit under the result, all calculated from the text in the box:
- Words — hyphenated words count as separate words, so
follow-upis two. - Characters — grapheme clusters, so one emoji is one character.
- Sentences — split on full stops, question marks and exclamation marks. The splitter mishandles some abbreviations:
a.m.andp.m.are cut after the first full stop, which adds a sentence. - Speaking time — at 130 words per minute, a comfortable presentation pace.
- Reading time — at 200 words per minute.
- Filler words — a fixed list: um, uh, erm, like, you know, sort of and kind of.
The filler count is deliberately crude. It matches like wherever it appears, so "it looks like rain" adds one, and it counts a multi-word filler such as "you know" once rather than twice. Read it as a hint about how spontaneous the transcript sounds, not as an error count.
A worked example
The first loaded sample is a three-sentence meeting note. It produces 33 words, 191 characters, 3 sentences, 15 seconds of speaking time, 10 seconds of reading time and no filler words.
Notes from the planning meeting. The first milestone moves to the second week of March. Priya will send the revised schedule on Thursday, and Sam will check the budget before the next review.
Words:
33 Sentences: 3
Speaking:
15 sec Reading: 10 sec Fillers: 0Two names appear in that transcript and both are the kind of word recognition gets wrong. Check every proper noun against what you meant to say before you send the note. A short filler test shows the other side of the count: Um, so I like think that, you know, it is sort of done. Uh, kind of. is 16 words, 2 sentences and 6 filler words — accurate about the ums, and one of those six is the ordinary verb "like".
Limits
- No audio files. The page listens to the live microphone only. The upload button accepts
.txtand loads a text transcript; it will not open an MP3 or a WAV. - No timestamps or speaker labels. You get one continuous block of text, so a two-person interview needs marking up by hand.
- No punctuation guarantee. Some engines accept spoken commands such as "full stop" or "comma"; support varies by browser, so test it before relying on it. Quotation marks and paragraph breaks are safer added in the edit.
- Ten locales only. A language outside the list cannot be selected even if the browser supports it.
- Quality depends on the room. A headset in a quiet room beats a laptop microphone near a conversation every time.
Where the audio goes
ToolsHub4u does not receive your audio or your transcript. The recognition request is made by the browser, from this page, to whichever speech service the browser vendor uses: Chrome may send audio to Google, Edge to Microsoft, Safari to Apple. Whether that processing happens on a device or on a remote server is decided by the browser and the platform, and this page cannot tell you which you got.
If the audio must not leave the device, do not use browser dictation. The transcript itself is ordinary page text: it is not saved, synchronised or sent anywhere by this site, and closing the tab discards it. Copy or download it first.