Verbal transcription software, while the meeting is under way
Hand-written transcripts lost details, and whoever wrote them missed the meeting.
Public administration
Software for public administration, on-premise
Real-time speech-to-text for the meetings of a public administration in Emilia-Romagna, with speaker diarisation, installed on-premise on the organisation’s own infrastructure.
Projects in this sector
How can a public administration use speech recognition without its data leaving?
By running it in-house. For a public administration in Emilia-Romagna we wrote software that transcribes meetings, live and afterwards, and tells speakers apart: the audio does not leave the client’s infrastructure, the speech recognition model runs there, and no third-party service sits between the recording and the final text.
Recurring problems in this sector
- Minutes transcribed by hand, losing detailsWhoever writes the minutes while the meeting is going on makes mistakes, forgets, misreads a tone. And meanwhile does not follow the discussion they should be taking part in.
- Audio that cannot leave the organisationA recorded meeting contains names, decisions, personal data. Sending it to an external speech recognition service means taking it outside the organisation’s infrastructure.
- Knowing who said whatWhen many people speak, the minutes have to say who spoke and when, and take the pauses into account.
- Minutes that arrive while the meeting is still freshThose who were there want them straight away; those who were not want to be able to listen again to the passage they care about.
- Background noise and unclear wordsIn a meeting room there is background noise and voices overlap, and a badly transcribed text has to be corrected word by word.
What we built
A platform that follows the meeting from start to finish: the invitations, the heading of the document before starting, the recording, live and deferred transcription, the follow-ups. While people talk the text is written, with the speakers told apart from one another; it arrives straight away by email to all participants, and the recording stays, to listen to again.
Speech recognition is a machine learning model trained for speech to text, the automatic conversion of voice into text. It removes background noise, reads pauses as punctuation and, during dictation, checks spelling and flags unclear words to whoever proofreads. Two months, with two backend developers and one frontend developer.
For a public university we also built a virtual microscopy platform, on which teachers and students look at tissue images together.
Where the software and the data run
On the organisation’s infrastructure: that is what on-premise means. The software and the speech recognition model are installed there, and from there the platform talks to the office tools: email, to send the text to those who took part, and the systems already in use, which it can integrate with.
What to ask a supplier
Before entrusting a supplier with software that listens to the organisation’s meetings, ask:
- Where does the software run, and where does the audio go while it is being transcribed?
- Which third-party services does the data pass through, from recording to text?
- Can the text be corrected, and who flags the words the system did not understand?
- Which of the organisation’s tools does it integrate with?
- Is the supplier certified for information security, and by which body?
- Who owns the code and the licences at the end of the work?
What a public body should know before having software built →
After a first call, free of charge, we work out with whoever runs the organisation’s information systems where the data and the software must live, and which tools it has to talk to.
The scope is written in phases, and for each phase we state the date and what it delivers.
Frequently asked questions
Can we have the transcript while the meeting is under way?
Yes, and also once the meeting is over, afterwards. Those who took part receive the text by email without waiting.
Does the software tell who is speaking?
Yes. It keeps track of contributions person by person, and reads pauses as punctuation.
Can the text be corrected?
Yes. Whoever proofreads has the recording to check a passage, and the software shows them where to look, because it flags the words it did not understand.
Is dotenv ISO 27001 certified?
Yes: UNI CEI EN ISO/IEC 27001:2024, certificate IIS-1225-06, issued by Dasa-Räegister under ACCREDIA accreditation. It is the standard for managing information security.
How long does a project like this take?
The transcription one took two months, with a team of three people. A wider project takes longer, and is delivered in phases, each with a date.
Tell us about the project you have in mind.