Hybrid method for identification of speech signals acquired by a distributed fiber optic sensor

Abstract

Distributed Acoustic Sensing (DAS) has emerged as a promising technology for monitoring and information-security applications due to its ability to provide continuous distributed sensing over long distances using standard optical fibers. However, speech signals acquired by DAS systems are characterized by low signal-to-noise ratios, spectral instability, environmental interference, and significant distortions, which complicate automatic speech recognition and speaker identification. This paper presents a hybrid approach for the processing and identification of speech signals acquired by a distributed fiber-optic acoustic sensor. The proposed framework combines Probability Difference-Based Temporal Fusion (PDTF), Discrete Wavelet Transform (DWT), Mel-Frequency Cepstral Coefficients (MFCC), Common Spatial Patterns (CSP), and Support Vector Machine (SVM) classification. A review of contemporary methods for DAS-based speech processing is conducted, and the role of time-frequency analysis, perceptually motivated feature extraction, temporal fusion, and machine-learning techniques is discussed. The proposed framework integrates temporal, spectral, perceptual, and spatial information within a unified processing pipeline designed to improve the robustness of speaker identification under noisy operating conditions. The analysis indicates that hybrid signal-processing architectures provide significant potential for DAS-based speech recognition by combining the advantages of classical signal-processing methods and machine-learning algorithms. The presented approach may be applied in information-security systems for covert monitoring, personnel authentication, intruder detection, perimeter protection, and critical-infrastructure security. Future research directions include adaptive noise-suppression techniques, self-supervised learning methods, multimodal data fusion, and the development of specialized benchmark datasets for DAS-based speech-recognition systems.