Published Online:September 2026
Product Name:The IUP Journal of Information Technology
Product Type:Article
Product Code:IJIT030926
DOI:10.71329/IUPJIT/2026.22.3.46-59
Author Name:Manisha Jangra, Ritu Boora, Vijay Kumar, Abhimanyu Nain and Ajay Kumar
Availability:YES
Subject/Domain:Engineering
Download Format:PDF
Pages:46-59
Binaural sound source localization (BSSL) is widely used in intelligent robots for effective human computer interactions and environmental awareness. Its primary objective is to accurately estimate the direction of arrival (DOA) of a sound source, allowing systems to orient and track speakers in real-time. These systems seek to replicate the human hearing system using dual microphones in humanoids. However, the real-application scenarios pose several challenges to these systems. Hence, this paper presents a dual-channel convolutional neural network (CNN)-based method for BSSL to narrow the research gaps in this area. The hierarchical structure in CNN extracts the individual channel features in the early stage and progressively exploits the relational information between the paired signals. The proposed method attained 95.29% accuracy with a 0.1 learning rate and a loss of 0.0691 when evaluated on the binaural dataset. It shows superlative performance when compared to the baseline methods based on CNN and fusion of CNN and deep neural network (DNN). The experimental results demonstrate the effectiveness of the method in estimating the DOA in 2D.
Binaural sound source localization (BSSL) estimates the position of a sound source with respect to a single pair of microphones. It is influenced by principles underlying the human auditory perception system. Over the last decade, it has garnered significant attention from researchers because of its critical role in modern applications.