PGA 2025.08.20

This article lists the new updates that are now available to all Pharia Government Assistant customers.


Transcription App (beta)

  • High quality audio and video transcription generated with the Whisper Large v2 model by OpenAI (Open Source). Only 4-5% errors per word (WER) for German in benchmarks. Supports a wide range of languages.

  • Easy speaker management: Automated voice-based speaker distinction and segmenting. Listen to short audio examples to name participants. Add other participants for referencing in the minutes.

  • Upload files up to 300 MB as mp3, mp4, mpeg, mpga, m4a, wav and webm. This represents up to 5 hours of mp3 audio and up to 30 minutes of Microsoft Teams Recordings (mp4) or uncompressed wav.

  • Transcription task management: Work on multiple transcription tasks at the same time. Search, delete, and rename transcription tasks.

  • Export transcripts to Word (docx) files for further processing. Alternatively, copy them to the Chat app to create summaries and structured minutes.

Increased File Upload in Chat

The file upload capabilities have been increased to 10 files, where each file can be up to 500 pages or 100 MB. Learn more…​

Improved Explainability in Chat

  • In the sources feature, the clarity of the references list has been improved, and it has been structured by document.

  • Contradicting information in sources is now shown separately.

The sources feature is unavailable for messages that do not have any documents and summaries.

Revamped Chat App

The Chat user interface has been improved. Changes include the following:

  • A new, more user-friendly design.

  • A new option to regenerate an answer in case of errors.

  • A new stop button to terminate the model’s thinking process to be able to correct errors without having to wait for the entire answer. Learn more…​

  • The maximum prompt size has been increased to 80,000 characters. (Note that this is independent of the model’s context window of 128,000 tokens.)