Deep Convolutional Network for Animal Sound Classification and Source Attribution using Dual Audio Recordings

Oikarinen, Tuomas; Srinivasan, Karthik; Meisner, Olivia; Hyman, Julia; Parmar, Shivangi; Desimone, Robert; Landman, Rogier; Feng, Guoping

doi:10.1101/437004

Cited by 7 publications

(13 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…IOR was 0.86 for whether a call occurred and 0.91 for call type. This data was then used to train a neural network for auto-detection [26]. The remaining two pairs were annotated with the neural network and corrected by one observer.…”

Section: Discussionmentioning

confidence: 99%

Close-range vocal interaction in the common marmoset (Callithrix jacchus)

et al. 2020

Self Cite

View full text Add to dashboard Cite

Vocal communication in animals often involves taking turns vocalizing. In humans, turn-taking is a fundamental rule in conversation. Among non-human primates, the common marmoset is known to engage in antiphonal calling using phee calls and trill calls. Calls of the trill type are the most common, yet difficult to study, because they are not very loud and uttered in conditions when animals are in close proximity to one another. Here we recorded trill calls in captive pair-housed marmosets using wearable microphones, while the animals were together with their partner or separated, but within trill call range. Trills were exchanged mainly with the partner and not with other animals in the room. Animals placed outside the home cage increased their trill call rate and uttered more trills in response to their partner compared to strangers. The fundamental frequency, F0, of trills increased when animals were placed outside the cage. Our results indicate that trill calls can be monitored using wearable audio equipment and that minor changes in social context affect trill call interactions and spectral properties of trill calls.

show abstract

Section: Discussionmentioning

confidence: 99%

Close-range vocal interaction in the common marmoset (Callithrix jacchus)

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…As shown, the default setting applies some amount of over-and undersampling, while not completely balancing distributions. The reasons were (a) decreasing training time for the initial experiments by reducing the data set size, (b) ensuring that the extreme imbalance in the training set does not prevent training convergence [19], and (c) to imitate the default setting commonly used in animal call detection systems, which usually start with already undersampled databases [10,12,14].…”

Section: Resamplingmentioning

confidence: 99%

“…In the annual DCASE-challenges, deep-learning based systems became popular between 2016 and 2017 [6][7][8][9]. Research teams working on automatic animal call detection particularly adapted convolutional neural networks (CNNs) with spectrogram inputs: Bergler et al [10] applied Res-Net [11] variants for detection of orca calls in long-term recordings; Bjorck et al [12] applied Dense-Net CNNs [13] for detecting African forest elephants with PAM; Oikarinen et al [14] applied siamese CNNs with stereo inputs to the detection of various marmoset monkey calls. Aodha et al [15] investigated CNNs for bat detection.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Compensating class imbalance for acoustic chimpanzee detection with convolutional recurrent neural networks

Anders¹,

Kalan²,

Kühl³

et al. 2021

Preprint

View full text Add to dashboard Cite

Automatic detection systems are important in passive acoustic monitoring (PAM) systems, as these record large amounts of audio data which are infeasible for humans to evaluate manually. In this paper we evaluated methods for compensating class imbalance for deep-learning based automatic detection of acoustic chimpanzee calls. The prevalence of chimpanzee calls in natural habitats is very rare, i.e. databases feature a heavy imbalance between background and target calls. Such imbalances can have negative effects on classifier performances. We employed a state-of-the-art detection approach based on convolutional recurrent neural networks (CRNNs). We extended the detection pipeline through various stages for compensating class imbalance. These included (1) spectrogram denoising, (2) alternative loss functions, and (3) resampling. Our key findings are: (1) spectrogram denoising operations significantly improved performance for both target classes, (2) standard binary cross entropy reached the highest performance, and (3) manipulating relative class imbalance through resampling either decreased or maintained performance depending on the target class. Finally, we reached detection performances of 33 % F 1 for drumming and 5 % F 1 for vocalization, which is a > 7 fold increase compared to previously published results. We conclude that supporting the network to learn decoupling noise conditions from foreground classes is of primary importance for increasing performance.

show abstract

“…In the field of audio classification and speech recognition, to the best of our knowledge, this is the first library specifically designed for audio data augmentation. Audio data augmentation techniques fall into two different categories, depending on whether they are directly applied to the audio signal [5] or to a spectrogram generated from the audio signal [6]. We propose 15 algorithms to augment raw audio data and 8 methods to augment spectrogram data.…”

Section: Introductionmentioning

confidence: 99%

Audiogmenter: a MATLAB toolbox for audio data augmentation

Maguolo

Paci

Nanni

et al. 2021

ACI

View full text Add to dashboard Cite

PurposeCreate and share a MATLAB library that performs data augmentation algorithms for audio data. This study aims to help machine learning researchers to improve their models using the algorithms proposed by the authors.Design/methodology/approachThe authors structured our library into methods to augment raw audio data and spectrograms. In the paper, the authors describe the structure of the library and give a brief explanation of how every function works. The authors then perform experiments to show that the library is effective.FindingsThe authors prove that the library is efficient using a competitive dataset. The authors try multiple data augmentation approaches proposed by them and show that they improve the performance.Originality/valueA MATLAB library specifically designed for data augmentation was not available before. The authors are the first to provide an efficient and parallel implementation of a large number of algorithms.

show abstract

Deep Convolutional Network for Animal Sound Classification and Source Attribution using Dual Audio Recordings

Cited by 7 publications

References 22 publications

Close-range vocal interaction in the common marmoset (Callithrix jacchus)

Close-range vocal interaction in the common marmoset (Callithrix jacchus)

Compensating class imbalance for acoustic chimpanzee detection with convolutional recurrent neural networks

Audiogmenter: a MATLAB toolbox for audio data augmentation

Contact Info

Product

Resources

About