Introduction

DCASE 2019 Challenge will offer awards for open-source and innovative methods. These awards are meant to encourage open science and reproducibility, and therefore the Reproducible System Award is directly based on these criteria. In addition, through our Judges’ Award we want to encourage novel and innovative approaches.

Each task will offer two awards (except Task 2 that offers prizes through Kaggle). These awards aim to encourage participants to openly publish their code, and to use novel and problem-specific approaches, and are therefore not directly based on the evaluation set performance ranking. We also highly encourage student authorship.

Awards

Open source Award

Reproducible system award

Reproducible system award of 500 USD will be offered for the highest scoring method that is open-source and fully reproducible. For full reproducibility, the authors must provide all the information needed to run the system and achieve the reported performance. The choice of licence is left to the author, but should ideally be selected among the ones approved by the Open Source Initiative.

Award recipients


Scenes Task 1	Mark D. McDonnell and Wei Gao Acoustic Scene Classification Using Deep Residual Networks with Late Fusion of Separated High and Low Frequency Paths	PDF
Localization Task 3	Yin Cao, Turab Iqbal, Qiuqiang Kong, Miguel B. Galindo, Wenwu Wang, Mark D. Plumbley Two-stage sound event localization and detection using intensity vector and generalized cross-correlation	PDF
Domestic Task 4	Liwei Lin, Xiangdong Wang, Hong Liu, and YueLiang Qian Guided Learning Convolution System for DCASE 2019 Task 4	PDF
Urban Task 5	Sainath Adapa Urban Sound Tagging Using Convolutional Neural Networks	PDF

Judges Award

Judges’ award

Judges’ award of 500 USD will be offered for the method considered by the judges to be the most interesting or innovative. Criteria considered for this award include but are not limited to: originality, complexity, student participation, open-source, etc. Single model approaches are strongly preferred over ensembles; occasionally, small ensembles of different models can be considered, if the approach is innovative.

Award recipients


Scenes Task 1	Paul Primus and David Eitelsebner Acoustic Scene Classification with Mismatched Recording Devices	PDF
Localization Task 3	Jingyang Zhang, Wenhao Ding, and Liang He Data Augmentation and Prior Knowledge-based Regularization for Sound Event Localization and Detection	PDF
Domestic Task 4	Teck KaiChan, Cheng Siong Chin, and Ye Li Non-negative Matrix Factorization-convolution Neural Network (NMF-CNN) For Sound Event Detection	PDF
Urban Task 5	Daniel Tompkins and Eric Nichols DCASE 2019 Challenge Task 5: CNN+VGGISH	PDF