Marine mammal sound anomaly and quality detection using multitaper spectrogram and hydrophone big data

This paper proposes a novel method for anomaly and quality detection of marine mammal sounds using multitaper spectrogram and hydrophone big data. The proposed method is aimed to automatically detect anomaly, such as high-frequency vessel noise, Doppler noise, in sperm whale (SPW) sound as well as the quality of the sound. A new signature function derived from a multi-taper spectrogram is able to detect the anomaly in the data and a new anomaly distortion measure can detect the sound quality into good/bad. The proposed method, is tested with 1905 minutes of data spanning a single year, and using a human operator's annotations. The experimental results reveal that the proposed multitaper spectrogram based approach is efficient in detecting anomaly as well as sperm whale sound quality for hydrophone big data and high detection accuracy (>85%) is achieved for raw input hydrophone data.