@techreport{Anderson2021,
    Author = "Anderson, Mark and Harte, Naomi",
    title = "Bioacoustic Event Detection with Prototypical Networks and Data Augmentation",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "This report presents deep learning and data augmentation techniques used by a system entered into the Few-Shot Bioacoustic Event Detection for the DCASE2021 Challenge. The remit was to develop a few-shot learning system for animal (mammal and bird) vocalisations. Participants were tasked with developing a method that can extract information from five exemplar vocalisations, or shots, of mammals or birds and detect and classify sounds in field recordings. In the system described in this report, prototypical networks are used to learn a metric space, from which classification is performed by computing the distance of a query point to class prototypes, classifying based on shortest distance. We describe the architecture of this network, feature extraction methods, and data augmentation performed on the given dataset and compare our work to the challenge's baseline networks"
}

@techreport{Bielecki2021,
    Author = "Bielecki, Radoslaw",
    title = "FEW-SHOT BIOACOUSTIC EVENT DETECTION WITH PROTOTYPICAL NETWORKS , KNOWLEDGE DISTILLATION AND ATTENTION TRANSFER LOSS",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "The report presents the results of submission to Task 5 (Few-shot Bioacoustics Event Detection) of Detection and Classification of Acoustic Scenes and Events Challenge (DCASE) 2021. This task focuses on sound event detection in a few-shot learning setting for animal (mammal and bird) vocalizations. Main issue of this task is the very limited number of training instances. The presented approach is based on prototypical networks built up from the convolutional layers. Main techniques used during model development are knowledge distillation, attention transfer loss and spectrogram augmentation. The best of presented models achieved 55.5\% F-measure on the challenge validation set. That is improvement by over 10\% in comparison to baseline model."
}

@techreport{Cheng2021,
    Author = "Cheng, Hao and Hu, Chenguang and Liu, Miao",
    title = "PROTOTYPICAL NETWORK FOR BIOACOUSTIC EVENT DETECTION VIA I-VECTORS",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "In this technical report, we present our system for the task 5 of Detection and Classification of Acoustic Scenes and Events 2021 (DCASE2021) challenge, i.e. few-shot bioacoustic event detection. First, per-channel energy normalization (PCEN) and i-vectors are extracted as features. In order to improve the diversity of original audio, some data augmentation methods are adopted, for example, specaugment. Then, the prototypical network with convolutional neural networks (CNN) is used for few-shot detection. Finally, we use aforementioned features as inputs to train our CNN model. We evaluate the proposed systems with overall F-measure for the whole of the evaluation set, and our best F-measure score on the validation set is 46.28."
}

@techreport{Johannsmeier2021,
    Author = "Johannsmeier, Jens and Stober, Sebastian",
    title = "FEW-SHOT BIOACOUSTIC EVENT DETECTION VIA SEGMENTATION USING PROTOTYPICAL NETWORKS",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "This report describes our submission to task 5 of the 2021 DCASE challenge. We detail how we processed the data, the model structure as well as the training procedure. We may submit an extended version to the DCASE 2021 workshop."
}

@techreport{Tang2021,
    Author = "Tang, Tiantian and Liang, Yunhao and Long, Yanhua",
    title = "TWO IMPROVED ARCHITECTURES BASED ON PROTOTYPE NETWORK FOR FEW-SHOT BIOACOUSTIC EVENT DETECTION",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "In this technical report, we describe our submission system for DCASE2021 Task5:few-shot bioacoustic event detection. Few improvements are investigated to better the baseline of deep learn- ing prototypical network. Including the N-way 5-shot classification prototypical network training strategy, data augmentation techniques, the proposed embedding propagation and attention similarity approaches. On the official validation set, we demonstrate that the proposed method achieves the overall F-measure score of 54.7\% on the validation set."
}

@techreport{Zhang2021,
    Author = "Zhang, Yue and Wang, Jun and Zhang, Dawei and Deng, Feng",
    title = "FEW-SHOT BIOACOUSTIC EVENT DETECTION USING PROTOTYPICAL NETWORK WITH BACKGROUND CLASSs",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "Few-shot bioacoustic event detection is a task to detect and classify bioacoustic events with only a few instances. This task was firstly introduced in DCASE2021 Task 5, which requires participants to create a method that can extract information from five sample sounds (shots) of mammals or birds, and detect sounds in field recordings. In this paper, a prototypical network-based method was proposed for few-shot bioacoustic event detection challenge. In order to detect the target event from the query sequence, we need to distinguish the target event, other events, and background noise with only a few support set. To solve this problem, we propose to sample background noise from the training dataset as the ”NEG” class for small sample learning. To better distinguish between events and background noise, the ”NEG” class is used as a ”way” in each episode of training. Experimental results show that the proposed method can effectively distinguish target events and background noise. The F-measure of sound event detection(SED) in the DCASE2021 Task 5 dataset can reach 57.10\%, which is higher than the baseline method(41.48\%)."
}

@techreport{Zou2021,
    Author = "Yang, Dongchao and Wang, Helin and Ye, Zhongjie and Zou, Yuexian",
    title = "FEW-SHOT BIOACOUSTIC EVENT DETECTION = A GOOD TRANSDUCTIVE INFERENCE IS ALL YOU NEED",
    institution = "DCASE2021 Challenge",
    year = "2021",
    month = "June",
    abstract = "In this technical report, we describe our few-shot bioacoustic event detection methods submitted to Detection and Classification of Acoustic Scenes and Events Challenge 2021 Task 5. We analyze the reason why Prototypical networks cannot perform well, and propose to use transductive inference for few shot learning. Our method maximizes the mutual information between the query features and their label predictions for a given few-shot task, in con- junction with a supervision loss based on the support set. Furthermore, we propose a mutual learning framework, which makes feature extractor and classifier to help each other. Experimental results indicate our transductive inference method get better performance than baseline, and F1 score is about 50.8\% on evaluation set. Furthermore, our mutual learning framework brings about 5\% improvement over the transductive inference method. We will release our code on https://github.com/yangdongchao/ DCASE2021Task5."
}