Turn a meeting recording into minutes

A recording becomes a document you can read, edit and circulate. Transcription happens on this device, and the transcript is shown to you before anything is saved.

• Paragraph breaks come from the silences in the recording itself.
• Save as PDF, Word or plain text, with timecodes if you want them.
• The audio never leaves the machine it was opened on.

The recording says where the paragraphs go

A transcript joined into one string is right for a search index and unreadable on a page: half an hour of talking is about four thousand words in a single block. The breaks are taken from the recording instead, because a pause is what a paragraph break is in speech, and every segment comes back with a start and an end.

Six tenths of a second is the threshold, which sits above ordinary breath between sentences and below the pause at the end of a thought. A speaker who never pauses is still broken up, by length and at a sentence boundary, because a lecture read off a script would otherwise come back as the wall this exists to prevent.

What to check before you circulate it

This is a recogniser and not a typist. It does not separate speakers, so the minutes will need attributions added, and only somebody who was in the room can confirm the names came out right. The transcript is shown on screen before anything is saved for exactly that reason.

Names are where it fails most often. A depot named twice in one recording came back wrong both times, while both of the figures in the same recording survived intact, so a place, a person or a company in these minutes is the thing to check first. Long spoken numbers are the other one: a phone number dictated aloud came back with two digits missing and the grouping rewritten as a decimal point, which reads like a figure somebody meant. Neither can be repaired, so each figure is counted and pointed at instead.

A recording with nothing said in it is refused by name rather than saved as an empty document, and one that comes back as a single short line is flagged, because that is the shape a recogniser makes out of silence.

Which audio formats work?

Whatever your browser can decode, which covers m4a, mp3, wav and most things a phone or a meeting tool produces.

How long can the recording be?

Up to half an hour per file. That is a budget in minutes rather than megabytes, because a megabyte of wav is six seconds and a megabyte of m4a is ten minutes.

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

The tool opens here once the page has finished loading. It needs JavaScript; the text above does not.