How text to speech works here
Type or paste text, pick a voice, and press Generate: the text is spoken by the same neural voices Microsoft Edge uses for Read Aloud and comes back as an MP3 you can play or save. There are 322 voices across 75 languages and 142 regional variants, Bangla among them.
Hello there. [pause 2s] This is after two seconds. [pause 0.5s] And a short one. → an 8.7-second, 51 KB MP3 in the Jenny voice, with the silences where the markers were
The one thing to know before you start: the text leaves your browser. It goes through this site's server to Microsoft's service, which does the speaking, and this site keeps no copy of it or of the audio. Documents opened on the File to Speech tab are read in your browser, and only the text you then generate from is sent.
Choosing a voice
Pick a language first, then a voice in it. The language list is grouped — English, then which English — because one language sounds different in each region: English (India) and English (United Kingdom) differ in accent, rhythm and how they read numbers and dates. When you switch region the tool keeps the same gender where the new region has one, so comparing accents is one click.
Each voice is trained on its own language and reads others poorly, if at all. The voices marked Multilingual — such as en-US-AndrewMultilingualNeural — are the exception: they follow the text from one language into another, which makes them the choice for a script that mixes languages.
Adding pauses
The Pauses menu inserts a marker at the cursor — [pause 1s] — and you can type one yourself: any value from 0.1 to 10 seconds, as in [pause 2.5s], up to 50 per generation. A marker outside that range is pointed out under the box rather than quietly changed.
A pause is added on top of the voice's own breath at the end of a phrase, which runs to about a second. A 2-second marker after a full stop therefore leaves a gap of roughly three seconds; for a tighter gap, use a smaller value.
Markers are plain text, so they survive copying and pasting, and they are never read aloud. Something that only looks like one — [pause] with no number — is read out as written.
Translating before speaking
Translate turns the whole box into the language of the voice you picked: choose an Arabic voice, press Translate, and English text becomes Arabic, ready to be spoken. The source language is detected for you, and nothing is translated until you press the button.
Pause markers are kept exactly as they were — Google Translate would otherwise translate [pause 2s] into words — and so are blank lines. Undo translation puts your original back until you edit the text. Machine translation is a draft, not a finished script: read it over before generating, especially names, numbers and anything formal.
The request goes from your browser to Google Translate's free web endpoint, not through this site. Google limits how much one network may send, and shared connections are sometimes refused outright; when that happens the tool says so and nothing in the box changes.
Every control
Each control on the page, what it does, and when it is worth reaching for.
| Control | What it does | Reach for it when |
|---|---|---|
| Text box | Holds up to 10,000 characters, markers included. The counter above it shows how many are used; past the limit, Generate turns off and the box says by how much. | Always — it is what gets spoken. |
| Pauses | Inserts [pause 0.5s] to [pause 5s] at the cursor. | A heading needs a beat after it, or a list needs room between items. |
| Clear text | Empties the box and the file notice under it. | Starting over. |
| Translate | Translates the box into the selected voice's language with Google Translate, keeping pause markers and line breaks. | Your text is in one language and the voice you want speaks another. |
| Language | Chooses the region, grouped by language. | You want a particular accent — Indian rather than British English, Bangladeshi rather than Indian Bangla. |
| Voice | Chooses a voice in that region. Each is listed with its gender; multilingual ones are marked. | Trying a different sound for the same text. |
| Rate | Speeds the voice up or slows it down, from −50% to +50% of its normal pace. | Narration for learners, slower, or a quick review, faster. |
| Volume | Raises or lowers loudness, from −50% to +50%. | The file will sit beside other audio and has to match it. |
| Pitch | Moves the voice up or down, from −50 Hz to +50 Hz. | A voice sounds a little too high or too low for the material. |
| Keep these settings | Remembers rate, volume and pitch in this browser. | You always use the same settings and want them back next visit. |
| Reset to default | Puts all three sliders back to zero. | The voice sounds odd and you want its natural delivery. |
| Generate | Sends the text to be spoken and shows progress as each part arrives. | The text and the voice are ready and the bot check has passed. |
| Stop | Abandons a generation in progress; nothing is kept. | You spot a typo halfway through a long script. |
| Download MP3 | Saves the result as an MP3 named after the voice and the time. | You want the audio outside this page. |
| Extract text | Reads a document in your browser and puts its text in the box on the other tab. | The words are already in a PDF, Word or spreadsheet file. |
Changing the text, the voice or a setting after a generation dims the result rather than removing it, so you can still play or save it — generate again to hear the change. Each press of Generate uses up the bot check, which then runs again by itself in a second or two.
Reading text out of files
The File to Speech tab opens a document in your browser, takes the words out, and puts them in the text box; the file itself is never uploaded. Up to 100,000 characters are taken from a document, and one generation reads up to 10,000, so you choose which part to send.
| Format | How it is read | Watch for |
|---|---|---|
.txt | As UTF-8, or as UTF-16 when the file starts with a byte-order mark. | Text saved in an older code page, such as Windows-1252, shows stray characters. |
.pdf | Page by page, with PDF.js — the engine behind Firefox's PDF viewer. | A PDF stores glyphs, and in Bangla, Hindi or Arabic they often come out in the wrong order. A scanned PDF holds pictures of text and gives nothing. |
.doc | The Word 97–2003 format, read from its own index of text pieces; the main body only. | Footnotes, headers and comments are left out. Files from Word 95 and older are not read. |
.docx | With Mammoth, paragraph by paragraph; tables cell by cell. | Pictures and charts are skipped, along with any text drawn inside them. |
.odt | From the document's own XML, paragraph by paragraph. | Notes and comments are left out. |
.xlsx, .xls, .ods | With SheetJS: every sheet, a line per row, cells joined by commas, values as the cell displays them. | A workbook of figures reads as a long list of figures — it is usually better to copy the cells you want. |
Limits
One generation reads up to 10,000 characters. Ten thousand characters of Bangla took about two minutes to generate and came back as fifteen minutes of audio, a 5.2 MB file, so a long text is best split into chapters.
Each address can start 30 generations an hour and the whole site 1,200. Every reader goes through this site's server, so Microsoft sees them all as one caller, and the second limit keeps one busy hour from spending that for everybody.
The voice service is not offered by Microsoft for other sites to build on. It is the service behind Edge's Read Aloud, reached the way Edge reaches it, and Microsoft changes it without notice. When it changes, this tool stops working until its client library catches up — and the error says so rather than failing quietly.
Where your text goes
Your text leaves the browser once, when you press Generate: it travels to this site's server, which passes it to Microsoft's speech service and streams the audio back. This site does not store the text or the audio and does not write either to its logs. The only record kept is a count for the hourly limit, under a salted hash of your address.
What Microsoft and Google do with text sent to them is governed by their terms, not this site's, so do not generate or translate speech from anything you would not paste into a website. Translate sends the text from your browser straight to Google Translate; documents opened on the File to Speech tab never leave your browser, and settings you choose to keep stay in your browser's storage.
Frequently asked questions
.docx or .odt and open that.<break> is refused outright, which is why pauses are markers in the text instead.