Section: New Software and Platforms

DCASE 2019 baseline

Baseline system for the task 4 of DCASE 2019 Challenge

Keywords: Audio signal processing - Audio source classification - Machine learning - Smart home

Functional Description: This is the baseline system for the task 4 of the challenge on Detection and Classification of Acoustic Scenes and Events (DCASE) 2019. The algorithm performs sound events detection and classification. From an audio recording, the target of the system is to provide not only the event class but also the event time boundaries given that multiple events can be present in an audio recording. The baseline approach relies on convolutionnal and recurrent neural networks (CRNN) and a mean-teacher model to exploit a large amount of unbalanced and unlabeled training data together with a small weakly annotated (without timestamps) training set and a strongly annotated (with timestamps) synthetic set to improve system performance.

Release Functional Description: This version includes a mean teacher model to exploit the various training sets with that have different levels of annotations, as provided in the task 4 of the DCASE 2019 challenge.

• Participants: Nicolas Turpault and Romain Serizel

• Contact: Nicolas Turpault

• Publication: Sound event detection in domestic environments with weakly labeled data and soundscape synthesis

• URL: https://github.com/turpaultn/DCASE2019_task4/tree/public/baseline