Deep metric learning for bioacoustic classification: Overcoming training data scarcity using dynamic triplet loss

Abstract

This paper proposes multiscale convolutional neural network (CNN)-based deep metric learning for bioacoustic classification, under low training data conditions. The proposed CNN is characterized by the utilization of four different filter sizes at each level to analyze input feature maps. This multiscale nature helps in describing different bioacoustic events effectively: smaller filters help in learning the finer details of bioacoustic events, whereas, larger filters help in analyzing a larger context leading to global details. A dynamic triplet loss is employed in the proposed CNN architecture to learn a transformation from the input space to the embedding space, where classification is performed. The triplet loss helps in learning this transformation by analyzing three examples, referred to as triplets, at a time where intra-class distance is minimized while maximizing the inter-class separation by a dynamically increasing margin. The number of possible triplets increases cubically with the dataset size, making triplet loss more suitable than the softmax cross-entropy loss in low training data conditions. Experiments on three different publicly available datasets show that the proposed framework performs better than existing bioacoustic classification frameworks. Experimental results also confirm the superiority of the triplet loss over the cross-entropy loss in low training data conditions.

Investigation of Different CNN-Based Models for Improved Bird Sound Classification

“…In the work of [30], a triplet sampling was used for generating the triplet spectrograms as the input to CNNs: full spectrogram, harmonic-component based spectrogram, and percussive-component spectrogram. Then, a dynamic triplet loss was used for the classification using multi-scale analysis module.…”

“…The duration of audio files in CLO-43DS data is different, which cannot be directly used as the input to the CNN. The first method for dealing with the multi-variate varying length audio data is that the signal is repeated from the beginning to force the fixed duration of 2s, which has been used in [30]. The second method is to directly resize the audio image to a fixed size.…”

Comparing recurrent convolutional neural networks for large scale bird species classification

“…However, it is a time-consuming and expensive endeavor to obtain a manually labeled dataset in bioacoustics, and it may also be very challenging to collect enough labeled data in practice, especially if a species rarely calls or if a species is rare. Given this scenario, some bioacoustics research works used other techniques in addition to CNN, including transfer learning with fine-tuning 36 – 39 , pseudo-labeling 40 , and using few-shot learning approaches 41 .…”

Recurrent Convolutional Neural Networks for Large Scale Bird Species Classification

“…However, it is a time-consuming and expensive endeavor to obtain a manually labeled dataset in bioacoustics, and it may also be very challenging to collect enough labeled data in practice, especially if a species rarely calls or if a species is rare. Given this scenario, some bioacoustics research works used other techniques in addition to CNN, including transfer learning with fine-tuning [28][29][30] , pseudo-labeling 31 , and using few-shot learning approaches 32 .…”