Comparison of Deep Learning Classifier Models for Sensor Array based Acoustic Source Localisation

Convolutional Neural Networks (CNNs) and Feed Forward Neural Networks (FFNNs) are used as classifier models for microphone array-based acoustic source localization. A microphone array captures spatial information of sound sources, enabling accurate localization through signal processing and deep learning techniques. A dataset comprising audio recordings from indoor environment, containing a single sound source was collected. The dataset was preprocessed to extract relevant features, and the localization ground truth was obtained using known source positions. Both CNN and FFNN models are implemented to learn the complex mapping between acoustic features and source locations. The CNN model leverages the inherent spatial relationships in the microphone array data through convolutional layers, capturing hierarchical representations of the input. On the other hand, the FFNN model processes the features sequentially through fully connected layers. Both the models are trained, validated, and tested using the data collected from a consistent experimental setup. The results demonstrate the comparative performance of CNNs and FFNNs in terms of localization accuracy, computational efficiency, and robustness to environmental variations. The strengths and limitations of each model are discussed and provide insights into their suitability for different acoustic source localization scenarios. This research contributes to the ongoing exploration of neural network architectures for improving the accuracy and efficiency of acoustic source localization systems. The findings offer valuable guidance for selecting appropriate models based on specific application requirements and computational considerations.