About

Political Debates is a project designed to facilitate the comprehensive analysis of video and audio content.

It streamlines the workflow from raw media to searchable analytics through three core capabilities:

AI-Powered Processing: Transcripts and translations are automatically generated and diarized (split by speaker).
Human-in-the-Loop: A dedicated user interface allows editors to annotate speakers and correct AI-generated text.
Dynamic Search: A specialized search engine updates in real-time with modified content to help locate specific speaker statements.

Context

This work was originally funded by the SNSF Spark Grant “Debating Human Rights”.

Grant Number: 221139 Source: SNSF Data Portal ⧉

The goal was to create specialized components for analyzing United Nations Human Rights Council (UNHRC) debates, focusing on:

Sophisticated Transcription: Integrating cutting-edge models (e.g., Whisper 3.0) for accurate, multilingual transcription.
Multimodal Handling: Processing video/audio for scene extraction and speaker diarization.
Specialized Database: Storing metadata and extracted features efficiently.

This repository consolidates the results of that project to make the processing tools available in a generalizable form.

Get started with the platform:

Proof of Concept (PoC)

This project is currently a Proof of Concept and is not production-ready. Please be aware of the following limitations:

Data Storage: Data is stored on Docker volumes limited by server disk space. There are currently no automated data recovery strategies in place.
Authentication: The system uses basic Nginx authentication to distinguish between two roles (Editor and Reader). There is no granular user management beyond these two roles.