Context
Transkripo is an artificial intelligence-powered software solution developed by ICT Linz for creating all types of minutes from audio files, offering practical tools for post-editing and refining transcriptions. With artificial intelligence, all speakers can be automatically recognised, even in different languages.
The Transkripo software solution:
- integrates seamlessly into the existing system landscape,
- requires no additional hardware – a smartphone is sufficient,
- provides a convenient export in MS Word, text or SRT formats,
- supports users with an improved workflow, allowing them to focus on what matters most.
The complete software stack including all AI components is hosted in-house in the data centre. Therefore, a GDPR-compliant, trustworthy, and secure environment for transcriptions can be offered where no data leaves the premises.
Objectives
In discussion with clients and experiments with first prototypes, a solution was implemented that offers up to 50 times faster transcription than the conventional (manual) way, thus saving time and costs.
Automated recognition of different speakers and supporting multilingual conversations and dialects was another objective. An integrated dictionary for technical terms, regional designations, and notable personalities enables optimal transcription results.
A web interface is used to upload the files and directly edit the transcription including adding names to the recognised speakers as well as a MS Word export of the transcription file.
Most importantly, the solution enhanced the digital sovereignty of the city by offering data protection-compliant hosting in Linz.
Implementation
ICT Linz started early to develop skills in hosting AI solutions on premise to be able to provide digital sovereignty for their clients: the administration of the City of Linz as well as several organisations owned by the city.
Initially, a VMware solution with NVIDIA cards was installed in the data centre to run Open AI’s Mistral and Whisper locally.
It was clear very early on that transcribing audio files, e.g. from city council or board meetings, is a use case where due to privacy and security concerns a future solution should not be built on free tools on the Internet, but need an on-premise solution. After first experimenting with an open-source solution from Switzerland, ICT Linz created an individual solution from scratch that is available as web-based service running in the data centre, and has since been used by customers on a regular basis.