Enhancing Accessibility of Parliamentary Video Streams: AI-Based Automatic Indexing Using Verbatim Reports

EasyChair Preprint 10892

Abstract

The increasing availability of documents and multimedia contents published by public Institutions and Administrations over the Internet pushes investments for improving their accessibility and navigability. The Italian Senate has been broadcasting video streams of its plenary sittings for the last two decades, but only since 2016 each video has been indexed according to the table of contents of the corresponding verbatim report, allowing citizens for accessing videos at the moment of each specific event indexed in the report. However, the elaboration of the augmented indexes necessary for achieving this kind of access requires a considerable effort. In this paper, we present a prototype system that automatizes the production of augmented video indexes for the plenary sittings not currently indexed. We exploit artificial intelligence technologies, such as Speaker Diarization and Speech2Text models, to transcript each sitting and cross-reference the results with sentences in the verbatim reports to create meaningful indexing files, named Video Table of Contents (VTOC). We evaluated our system against sittings of the 15th Italian term obtaining encouraging results.

Keyphrases: Semantic Textual Similarity, Speech2Text, Video Automatic Indexing, speaker diarization

@booklet{EasyChair:10892,
  author    = {Daniele Bertillo and Andrea De Donato and Carlo Marchetti and Paolo Merialdo},
  title     = {Enhancing Accessibility of Parliamentary Video Streams: AI-Based Automatic Indexing Using Verbatim Reports},
  howpublished = {EasyChair Preprint 10892},
  year      = {EasyChair, 2023}}