Context and motivations#
The DCLDE community is the place-to-be where the FAIR state-of-the-art of AI models for the detection and classification of marine mammal vocal sounds based on Passive Acoustic Montiroing ( PAM) should be developed. However, DCLDE has not yet fully succeeded in setting up such a benchmarking framework where best performing models could be easily found, reproduced and improved over time. The absence of well-defined reference tasks using standardized metadata, the lack of a single and centralized hosting platform, the difficulty of assessing quantitatively the quality of annotations, are among the reasons explaining this delay.
Objective#
Our long term objective is to develop ALL TOGETHER a FAIR and sustained data challenge (aka model benchmarking framework) for the PAM community. An initial focus has been put on the tasks of detection and classification of marine mammals sounds using a single hydrophone.
Methods#
On the road towards this ambitious objective, we have identified the following actions that will need to be undertaken :
- federating ongoing individual efforts through an official working group ;
- co-developing all materials (task, data, codes) to build the challenge ;
- interacting closely with ongoing international initiatives in PAM like Tethys and GLUBS ;
- drawing best practices from neighboring AI communities like DCASE ;
- setting up a permanent challenge organization (steering group, platform).
Preliminary achievements#
Our first works have been to: