As part of the European Defense Tech Hackathon 2025 in Munich, team Demfar developed audio2intelligence.
Given intercepted enemy radio communication, this tool extracts valuable intelligence, including:
- 🔐 Codewords
- 🧩 Speaker hierarchy
- 🗣️ Speaker descriptions
- ✍️ Text transcription
Note: You will need a Hugging Face and AssemblyAI token.
Open two terminals, one for the client and one for the server.
cd client
npm start
cd server
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python server.py
Next, create a .env file in the server directory.
Copy the content from .env.example, then fill in your Hugging Face and AssemblyAI tokens.
The client can be accessed by default at http://localhost:4200/.
The landing page shows all existing sessions and allows you to create new ones. Each session has its own isolated context, separate from others.
After opening the report for a session, the application displays the speaker hierarchy, session summary, extracted codewords, and uploaded audio files. It is also possible to upload further audio files by clicking the Add new audio button.
By clicking a node representing a speaker in the hierarchy graph, the speaker info pops up.
By pressing the View More button of an audio file, the speech transcription pops up. Here, the summary of the speech is displayed, along with a diarized transcription. The user can choose to view the transcription in English or Russian.
