Speech Enhancement Based on Array-Processing-Assisted Distributed Fiber Acoustic Sensing

Distributed fiber acoustic sensing (DAS) based on phase-sensitive optical time-domain reflectometry ( 8-OTDR) has attracted lots of attention due to its unique advantages. In security monitoring, speech recognition, and other relevant fields, DAS's ability to acquire speech signals is a key indicator for practical applications. However, the speech signal has broadband and nonstationary characteristics, so the conventional filtering is not sufficient, and the speech enhancement algorithm based on the characteristics of the speech signal is needed; in addition, the acoustic sensitivity of conventional fiber layout is not high enough, resulting in poor signal quality. In this article, speech enhancement based on array-processing-assisted DAS system is proposed to recover speech signal with high fidelity. Pulse compression 8-OTDR (PC 8-OTDR), combined with a multichannel speech enhancement algorithm based on optimally modified log-spectral amplitude (OM-LSA) is used. As far as we know, this is the first time that the multichannel OM-LSA speech enhancement is proposed for DAS. Compared with the signal before processing, the speech signal processed by this method has a great improvement in signal distortion, background noise intrusiveness, and signal overall quality. In addition, other signal enhancement methods, such as wavelet denoising (WD) and beamforming (BF) are also analyzed and compared. Our scheme outperforms in the above three aspects. This scheme does not need to obtain a large number of datasets in advance and can be transplanted to other DAS systems, which has great significance for realizing high-performance speech acquisition.