Aim
We are compiling the largest dataset of verified underwater sound clips available online to the scientific community for use in marine research and conservation. This portal is hosted by King Abdullah University of Science and Technology.
Why?
Having a large dataset (with a goal of containing over 100,000 short sound clips) will enable researchers to assess soundscapes. One way to use the repository is to create an automatic classification framework able to identify sounds in longer soundscape recordings around the world. A dataset contributed to by a range of ecosystems and habitats will allow both broader- and finer-scale analysis and comparisons.
What?
The final result will be an open online repository of sounds, allowing ongoing uploads and downloads of underwater sounds. All sounds are categorised with metadata allowing the user to filter by sound type, location, frequency range, and depth of hydrophone, and are credited to the recorder.
You
We're reaching out to you, because your position as a researcher will make your contribution absolutely invaluable, and be mutually beneficial to the community of acoustic researchers and conservationists alike. To make it as easy as possible, we have created a portal website where you can be invited to create a log-in, are given “contributor” status by one of our admins, and can upload your sounds and associated metadata. Even if you don't have the exact details, fill in as much information as possible and the sound will still be a valuable contribution. If you prefer, you can send sounds in a zip file to us with the accompanying details in a provided excel file.
Sound specifics
Sounds can be of biological, geological or anthropological origin, but must be recorded underwater in a marine or estuarine environment. Clips must be less than 1 minute and preferably more than 10 seconds. We aim to develop an automatic clip trimmer for longer clips but at the moment, clips must be shortened using your preferred software program (e.g. Audacity).
Why 1 minute or less?
The short nature of the clips allow easy and meaningful addition into machine learning algorithms. The diverse and complex statistical properties of sounds mean that identifying short clips is an easier task for machine learning to then re-discover within a soundscape. For example, a one hour dolphin recording will probably not exactly match with your new soundscape, however, using an array of short clips, it is much more likely to capture a spectral similarity and pick up that source.
In collaboration with a public outreach program, the Ocean Sound Bank
When you upload your sound, you have the option to select whether you want to contribute also to The Ocean Sound Bank. The OSB is a public engagement project being led by The Ocean Agency that has been endorsed by the UN Ocean Decade. It will run in parallel with the existing Ocean Image Bank. The idea is to provide a freely available library of ocean sounds for anyone to use in art, media or public engagement projects that support ocean conservation. We’re particularly interested in collating unusual, unknown and ‘unlikely’ ocean sounds that will engage people with the diversity and uniqueness of ocean life. All contributors will be credited in a similar way to the photographers who have contributed to the Ocean Image Bank. For more information, feel free to contact Tim Gordon or Richard Vevers.