From sound to story: GAS-Saudi’s graph-based solution for audio summarization in the deaf community

Abstract Providing coherent summarization for the deaf community can be challenging, particularly in capturing the details relationships between spoken content and user queries. To our knowledge, no prior attempts have been made to develop a graph-based framework to address this challenge. The curre...

Full description

Saved in:
Bibliographic Details
Main Authors: Raed Alharbi, Khalid Almalki
Format: Article
Language:English
Published: Springer 2025-07-01
Series:Journal of King Saud University: Computer and Information Sciences
Subjects:
Online Access:https://doi.org/10.1007/s44443-025-00051-0
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Providing coherent summarization for the deaf community can be challenging, particularly in capturing the details relationships between spoken content and user queries. To our knowledge, no prior attempts have been made to develop a graph-based framework to address this challenge. The current approaches often concentrate on extracting semantically relevant information from audio signals, which fails to consider the contextual coherence and logical flow between sentences. As a result, users are frequently presented with disjointed summaries that lack a structured narrative, making it difficult for them to follow the intended storyline. This paper presents GAS-Saudi, a proof-of-concept novel graph-based framework designed to enhance summarization for the deaf community by leveraging complex relationships within acoustic signals. The framework consist of several key components: (1) an audio extractor that transforms audio signals into textual representations, (2) a query-answer module that aligns audio content with user inquiries, and (3) a story learning module that constructs and analyzes sentence relationships through graph-based learning. We conduct empirical evaluations on a real-world dataset, demonstrating the effectiveness of GAS-Saudi in generating coherent and meaningful summaries. The empirical results indicate that the GAS-Saudi model outperforms state-of-the-art methods, achieving an accuracy of 91%. The experiment’s code can be found here: https://github.com/raed19/GAS-Saudi .
ISSN:1319-1578
2213-1248