User Guide
Choose your language first
Pick from 48 languages before you start — English, हिन्दी, বাংলা, தமிழ், తెలుగు, मराठी, ગુજરાતી, ಕನ್ನಡ, മലയാളം, ਪੰਜਾਬੀ, اردو and more, grouped so you can find yours quickly. The regional variants are not decoration: Bangla (Bangladesh) and Bangla (India) use different models, as do English (India) and English (US). Choosing the one that matches how you actually speak is the single biggest thing you can do for accuracy.
Press Start Dictating and allow the microphone
Your browser will ask permission the first time. Allow it and the button turns into Stop Dictating with a live status line. If you blocked the microphone earlier, the padlock icon in the address bar is where you undo that — the tool will tell you if that is what happened.
Speak naturally, and say your punctuation
Talk at a normal pace; there is no need to slow down or over-enunciate. Punctuation has to be spoken, because speech does not contain it:
“period” or “full stop” → . · “comma” → , · “question mark” → ? · “exclamation mark” → ! · “colon” → : · “semicolon” → ; · “new line” and “new paragraph” break the text.
These work in your own language too, not only in English. Dictating in Hindi, say “अल्पविराम” for a comma or “पूर्ण विराम” to end the sentence; in Bangla, “কমা” and “দাঁড়ি”; in Spanish, “punto y coma”; in Chinese, “句号”. Every one of the 48 languages has its own set, and English stays active alongside it because people genuinely mix the two mid-sentence.
The mark that lands suits the script rather than being forced into an English one. Hindi, Marathi, Bangla and Punjabi get the danda (।), Urdu gets ۔, Arabic gets ، ؛ ؟, and Chinese and Japanese get the full-width 。，？！ with no space in front, because those scripts do not separate words with spaces.
Sentences are capitalised for you after every full stop, question mark, danda and line break, so you never have to say “capital”.
Dictating a document that discusses punctuation, or a sentence like “add a comma there”? Untick Spoken punctuation below the transcript and every phrase is typed as an ordinary word instead.
Watch the grey line while you talk
Words appear in grey beneath the box before they settle into the transcript. That grey text is the engine’s current best guess, and it frequently rewrites itself as you finish a phrase — which is normal, not a fault. Only settled text is written into the box and counted.
Fix things without stopping
You can edit the transcript while dictation is still running; your correction is kept and speech continues from it. Two things worth knowing:
Click anywhere and the next words go there, not at the end — so you can drop a forgotten clause into the middle of a paragraph. Select a word and speak to replace it, which is the quickest way to fix a misheard name, and works even if you do not have that language’s keyboard installed.
Keep going through the pauses
Speech engines stop listening after a stretch of silence. The tool notices and restarts immediately, so a long dictation with thinking pauses keeps building instead of quietly ending. On phones each sentence is its own short session by design — you will not notice, but it is why mobile dictation is reliable here.
Take the text with you
Copy puts the whole transcript on your clipboard. Download .txt saves it as dictation.txt. Both handle every script correctly, so বাংলা or தமிழ் pastes into WhatsApp, Word or email intact. Clear empties the box when you want a fresh start — there is no undo, so copy first if you might want it back.
About the Speech to Text Tool
Most people type at roughly 40 words a minute and speak at around 150. That gap is the entire argument for dictation, and it is why voice typing wins for first drafts, long WhatsApp messages, meeting notes and anything you would rather think through out loud than compose on a keyboard.
What actually happens when you speak
This uses the Web Speech API built into your browser. Your microphone audio is streamed to the speech service your browser is wired to — Google’s, in Chrome and Edge — and the recognised text comes back. The recognition does not happen on your device. We are explicit about that because plenty of sites in this category imply otherwise: an internet connection is required, and this is not the tool for dictating anything genuinely confidential.
What we can tell you is where the text does not go. It is never sent to our servers, never stored, and never logged. It lives in the box on this page until you copy it, download it, or close the tab.
Why recognition gets some words wrong
A speech recogniser is not matching sounds to words one at a time. It runs two models together: an acoustic model that maps audio to plausible sounds, and a language model that weighs which word sequences are actually likely in that language. That combination is why context helps so much — and why an unusual name gets replaced with a common word that sounds similar. The system is choosing the more probable sentence, which is usually right and occasionally confidently wrong.
It also explains the three things that most improve your results: a closer microphone, since headset and phone mics beat a laptop across the room; less background noise, because competing speech is the hardest case of all; and full sentences rather than isolated words, because the language model has nothing to work with when there is no context around a word.
Why you have to say the punctuation
Speech genuinely does not contain punctuation. You signal it with pauses and intonation, and those cues are ambiguous enough that no mainstream recogniser infers them reliably. Saying “comma” and “new paragraph” feels strange for about two minutes and then becomes automatic — it is how professional dictation has always worked, long before computers were involved.
What most browser dictation tools then get wrong is assuming you say it in English. This one carries a phrase list for every one of its 48 languages — 289 phrases in all — and the list for your selected language is checked before the English one, with English still active behind it. That ordering is deliberate: a Spanish speaker saying “punto y coma” means a semicolon, and a naive matcher that tried “punto” first would produce a full stop followed by the stray word “y coma”. Phrases are therefore always matched longest-first, which is also what stops French “point d’interrogation” collapsing into a bare full stop.
Punctuation belongs to the script, not to English
A Hindi sentence does not end with a full stop; it ends with a danda, ।, and the same mark closes Marathi and Bangla sentences. Urdu ends with ۔ and asks with ؟. Arabic writes its comma as ، and its semicolon as ؛. Chinese and Japanese use full-width marks — 。，？！ — which are designed to occupy a full character cell, so putting a Latin space in front of one is visibly wrong to anyone reading it.
Those scripts also raise a subtler point: Chinese, Japanese, Thai and Khmer do not put spaces between words at all. A transcript assembled with English spacing rules comes out looking like “你好 世界 。” rather than “你好世界。”. The joining logic checks the selected language before deciding whether a space belongs there, which is a small detail that separates a transcript you can paste straight into a document from one you have to clean up first.
Why the regional variants matter
Language codes here follow BCP 47, the standard that distinguishes bn-BD from bn-IN and en-IN from en-US. These are not cosmetic labels. Each maps to a model trained on different speakers, so a Bangladeshi speaker gets noticeably better results from Bangla (Bangladesh) than from Bangla (India), and Indian English is recognised far better by English (India) than by English (US). If accuracy disappoints, changing this is the first thing to try.
Browser and device support, plainly
Chrome, Edge and Safari support speech recognition. Firefox does not ship a speech engine at all, so the tool will tell you rather than appear broken. On phones, the languages available depend on what your device’s speech engine has installed — if a language you picked returns nothing, that is a missing language pack on the device, not a fault here.
Who this is genuinely for
Students turning lecture thoughts into notes. Anyone whose typing is slower in their own script than their speech — which is most people writing Bangla, Hindi or Tamil on a phone. Writers getting a messy first draft down without the keyboard slowing the thinking. People with RSI, limited hand mobility, or a broken wrist. And anyone who has ever composed a long message while walking. New to dictation? Our voice typing guide covers the one language setting that fixes most accuracy problems, spoken punctuation in Hindi and English, and why dictation keeps stopping. For the real distribution of typing speeds, the WPM each job and exam actually requires, and a four-week plan that works, read what a good typing speed really is.
Frequently Asked Questions
Is this speech to text tool really free?
Yes, completely, with no sign-up, no time limit and no cap on how much you dictate. It uses the speech engine already built into your browser, so there is no API cost behind it to pass on to you.
Does it work in Hindi, Bangla and other Indian languages?
Yes. Hindi, Bangla (both India and Bangladesh), Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Urdu, Nepali and Sinhala are all supported, alongside 48 languages in total. Pick yours from the dropdown before you start — and choose the regional variant that matches how you speak, since it materially affects accuracy.
Why is speech to text not working on my phone?
Three usual causes. The microphone permission was denied — check the padlock icon in the address bar. The browser has no speech engine, which is the case in Firefox. Or the language you picked has no voice data installed on the device, which you fix in your phone’s settings under text-to-speech or voice input, not here.
How do I add commas and full stops?
Say them out loud. “comma”, “period” or “full stop”, “question mark”, “exclamation mark”, “colon”, “semicolon”, “new line” and “new paragraph” are all typed for you. Capital letters at the start of sentences are handled automatically, so there is no need to say “capital”.
Can I say punctuation in Hindi, Bangla or my own language?
Yes. Every one of the 48 languages has its own phrase list — 289 in total. In Hindi say “अल्पविराम” for a comma or “पूर्ण विराम” to end the sentence; in Bangla “কমা” and “দাঁড়ি”; in Tamil “காற்புள்ளி”; in Spanish “punto y coma”; in Chinese “句号”. English phrases keep working at the same time, so mixing “comma” into a Hindi sentence is fine — which matters, because the browser’s own recogniser often returns the English word anyway.
Does Hindi get a full stop or a danda?
A danda — । — because that is how Hindi, Marathi and Bangla actually end a sentence. The mark follows the script rather than being forced into an English one: Urdu gets ۔ and ؟, Arabic gets ، and ؛, and Chinese and Japanese get the full-width 。，？！. Sentence capitalisation understands all of them too, so the next sentence starts correctly.
Why is there no space before the full stop in Chinese?
Because Chinese, Japanese, Thai and Khmer do not separate words with spaces, and their punctuation is full-width — the mark already occupies a whole character cell. Inserting an English-style space would look wrong to any reader of those scripts, so the tool checks your selected language before deciding whether a space belongs there.
I want to type the word “comma”, not the symbol. Can I turn this off?
Yes — untick “Spoken punctuation” below the transcript and every phrase is typed as an ordinary word. Useful when dictating something that discusses grammar, or any sentence like “add a comma there”, where the feature would otherwise work against you. It stays on by default.
Is my voice recorded or stored anywhere?
Not by us — your audio and the resulting text never touch our servers, and nothing is logged. Your browser does send the audio to its own speech service to be transcribed, which is how the Web Speech API works in every browser that supports it. For genuinely sensitive material, use offline dictation software instead.
Does voice typing work without internet?
No. Recognition happens on your browser’s speech service rather than on your device, so a connection is required. If it stops working mid-sentence, a dropped connection is the first thing to check.
Why does it keep getting one particular word wrong?
Recognisers pick the most probable sentence, so unusual names and technical terms get swapped for common words that sound similar. The fastest fix is to select the wrong word and simply say the right one — it replaces the selection. Speaking in full sentences also helps, because the surrounding words give the model context to work from.
Can I edit the text while I am still dictating?
Yes. Correct anything at any time and your edit is kept, with dictation carrying on from it. You can also click anywhere in the text and the next words you speak will be inserted at that point rather than at the end, which is how you drop a missing clause into the middle of a paragraph.
Which browser works best for voice typing?
Chrome and Edge give the most consistent results, and Safari works well on Apple devices. Firefox does not include a speech engine, so the tool will say so rather than fail silently. On mobile, Chrome on Android and Safari on iPhone are both solid.
Can I dictate a long document?
Yes, there is no length limit. Speech engines stop after a stretch of silence, so the tool restarts them automatically and your transcript keeps building through pauses. Copy or download as you go if you are dictating something you cannot afford to lose to a closed tab.
