Listen to a document on your commute

A reading of the document, a summary of it, or an explanation of it, as an MP3 you can take with you. The voice runs on this device and the file never leaves it.

• Read in full, summarised, or explained at length. Three different jobs.
• Every number is spelled into words, because the voice cannot say digits.
• The length of the reading is estimated before you commit to it.

The voice cannot say seven, eight or nine

The model is character-level and carries its own alphabet, and a character outside it is not mispronounced, it is dropped before the model sees it. Measured one character at a time, the English voice keeps the digits nought to six and has no seven, eight or nine at all.

So a figure written as seven lakh eighty-nine thousand arrives as two noughts and is read "zero zero": a number that came out wrong, in silence, with nothing on screen to catch it. Every number is therefore spelled into words before synthesis rather than relying on whichever digits happen to survive, and the result is checked against the alphabet the model itself reports rather than against a constant.

Whether a run of digits is a quantity or an identifier decides how it is read: an account number is read digit by digit and a total is read as a total, in lakhs if that is how the page wrote it. Anything still unsayable is named on the panel rather than dropped quietly.

Three modes, and an estimate you can trust

Reading in full needs only the voice. The summary and the explanation add a language model on top, and both are written for the ear rather than the eye: no bullets and no headings, because a bullet read aloud is the word "dash".

Synthesis is linear in characters, so the length shown before you press the button is arithmetic rather than a guess. The rate was measured against whole documents rather than one line, which is why the estimate does not come out short.

What does the voice sound like?

One English voice, chosen because it is the one people will sit through. A second language was built and removed for that reason.

How large is the audio file?

You choose a bitrate and the size is exact rather than estimated, because a bitrate is a promise rather than a target.

Support and billing: support@gosmartpdf.com. Operated by Mrityunjay Kumar, India. Payments by Dodo Payments, our merchant of record.

Loading the editor…