A topic-specific representation learning framework for acoustic scene classification

doi:10.1016/j.asoc.2025.113588

A topic-specific representation learning framework for acoustic scene classification

2025-07-13

0

PRE

AI

DOI：10.1016/j.asoc.2025.113588

Original

Original request for help

Save

Abstract

En 中文

• A topic representation learning framework is proposed to model both intra- and inter-semantic dynamics for acoustic scenes. • A topic decoupling module is proposed to learn fine-grained features and model intra-semantic dynamics. • A graph neural network is employed to enhance interactions among topics, capturing inter-semantic dynamics across scenes. • We use topics as nodes of a graph to form a meaningful structure for information interaction in audio time-series data.

Keywords:

topic representation

intra-semantic dynamics

inter-semantic dynamics

graph neural network

acoustic scene classification

A topic-specific representation learning framework for acoustic scene classification

Abstract

Journal

Organization