Gaussian mixture model classification of odontocetes in the Southern California Bight and the Gulf of California

Abstract

A method for the automatic classification of free-ranging delphinid vocalizations is presented. The vocalizations of short-beaked and long-beaked common ͑Delphinus delphis and Delphinus capensis͒, Pacific white-sided ͑Lagenorhynchus obliquidens͒, and bottlenose ͑Tursiops truncatus͒ dolphins were recorded in a pelagic environment of the Southern California Bight and the Gulf of California over a period of 4 years. Cepstral feature vectors are extracted from call data which contain simultaneous overlapping whistles, burst-pulses, and clicks from a single species. These features are grouped into multisecond segments. A portion of the data is used to train Gaussian mixture models of varying orders for each species. The remaining call data are used to test the performance of the models. Species are predicted based upon probabilistic measures of model similarity with test segment groups having durations between 1 and 25 s. For this data set, 256 mixture Gaussian mixture models and segments of at least 10 s of call data resulted in the best classification results. The classifier predicts the species of groups with 67%-75% accuracy depending upon the partitioning of the training and test data.

Discrimination of individual tigers (Panthera tigris) from long distance roars

“…The HMM is a statistical state machine model used in nearly all human speech processing and recognition studies (Juang, 1984). In recent years, the use of HMMs for animal vocalization classification in species such as elephants and dolphins (Roch et al, 2007) has also achieved promising results. In essence, a HMM maps states in the model to a sequential pattern of acoustic observations, enabling calculation of a probabilistic match between the observation sequence and the underlying model.…”

Categorizing click trains to increase taxonomic precision in echolocation click loggers

“…Risso's dolphins click 58 trains contained peaks in energy at 22.4, 25.5, 30.5 and 38.7 kHz and at 22.2, 26.6, 33.7 and 37.3 59 kHz for white-sided dolphins. The spectral location was sufficient to discriminate between the 60 two species but site and instrument-specific anomalies reduced the confidence of the 61 classifications (Roch et al, 2007). In the same habitat, bottlenose dolphin and common dolphin 62 (Delphinus delphis) echolocation clicks were found to have a more uniform energy distribution 63 between 40 and 80 kHz (Soldevilla et al, 2008).…”

“…In the Pacific, the echolocation clicks of white-sided dolphins (Lagenorhynchus 56 obliquidens) and Risso's dolphins (Grampus griseus) have been shown to display consistent 57 peaks and notches in spectral energy below 48 kHz (Roch et al, 2007). Risso's dolphins click 58 trains contained peaks in energy at 22.4, 25.5, 30.5 and 38.7 kHz and at 22.2, 26.6, 33.7 and 37.3 59 kHz for white-sided dolphins.…”

Detection of Whales Using Deep Learning Methods and Neural Networks

“…These were then utilized for characterizing future shrieks by figuring the probability that a recorded shriek has a place with every class. Roch et al [29] utilized GMMs to decide the types of recorded dolphin shrieks. The recorded sign was part up into time allotments from which the cepstral coefficients were computed.…”

“…Discourse acknowledgment is an issue that has been widely examined in the past [24,26,30], and due to its comparability to our issue it is sensible to research how techniques for discourse acknowledgment can be connected to perceiving up calls. Roch et al [29] and Brown and Smaragdis [11] utilized a methodology exceptionally like the one that was proposed for discourse acknowledgment by Rabiner in 1989 [26]. They too utilized the Cepstral coefficients which are utilized frequently as a part of discourse acknowledgment on the grounds that it conveys much data about the vocal tract [23] tool kit Voicebox [7].…”