Detection strategy for long-term acoustic monitoring of blue whale stereotyped and non-stereotyped calls in the Southern Indian Ocean

Abstract

The most common approach to monitor mysticete acoustic presence is to detect and count their calls in audio records. To implement this method on large datasets, polyvalent and robust automated call detectors are required. Evaluating their performance is essential, to design a detection strategy adapted to study the available datasets. This assessment then enables accurate post-analyses and comparisons of multiple independent surveys. In this paper, we present the performance of a detector based on dictionaries and sparse representation of the signal to detect blue whale stereotyped and non-stereotyped vocalizations (D-calls) in a larg acoustic database with multiple sites and years of recordings in the southern Indian Ocean. Results show that recall increases with the SNR (Sound to Noise Ratio) and reaches 90% for positive SNR stereotyped calls and between 80% and 90% for high SNR D-calls. A detailed analysis of the influence of dictionary composition, SNR of the calls, manual ground truth as well as interference types and abundance, on the performance variability is presented. Eventually, a detection strategy for long term acoustic monitoring is defined.

Distribution of blue whale populations in the Southern Indian Ocean based on a decade of acoustic monitoring

“…A threshold corresponding to one false positive detection per hour was set and the corresponding recall reached 90% for positive SNR calls (Torterotot et al, 2019). The detector was then tested on an entire year of data ( 2015) at WKER site and every detection was manually double checked, to characterize any interferences that could fool the algorithm and to re-evaluate the false alarm rate in case of call absence.…”

“…The performance evaluation of the algorithm on the same data used in this paper is thoroughly discussed in Torterotot et al (2019).…”

Long‐term acoustic monitoring of nonstereotyped blue whale calls in the southern Indian Ocean

“…If the signal is well reconstructed by the linear combination of K elements among the dictionary's waveforms, then the resemblance metric is higher than the threshold measured empirically in Torterotot et al, (2019), meaning that it is likely that a D-call lays in the observed signal (Guilment et al, 2018;Socheleau et al, 2018). Before applying the SRD to detect D-calls on the whole data set, its performance was investigated on manually annotated data subsets containing 240.5 hr of recordings from different years and locations of the OHASISBIO data set (NEAMS 2015, SSWIR 2017, WKER 2015, and WKER 2012 with a total of 3,467 D-call annotations, to encompass potential temporal and geographical variation of the calls.…”

“…Although visual examination of the data proved to be the most reliable technique to detect D-calls, we had to resort to a semiautomated detection method to process such a large amount of data. We selected an automated detection algorithm based on dictionary learning and sparse representation (Socheleau et al, 2018) and combined it with a postprocessing algorithm to reject false positives (Torterotot et al, 2019).…”

Detecting, classifying, and counting blue whale calls with Siamese neural networks

“…To investigate the vocal repertoire of Southeast Alaskan humpback whales, three classification systems were used, including aural spectrogram analysis, statistical cluster analysis, and discriminant function analysis, to describe and classify vocalizations, and a hierarchical acoustic structure was identified to classify vocalizations into 16 individual call types nested within four vocal classes (Fournet et al, 2015). For blue whale signals in particular, most detection methods have been based on detection either in the time domain (e.g., matched filtering; Stafford et al, 1998) Socheleau et al, 2015;Torterotot et al, 2019).…”