3 ms·
So funny to see this! I've had the pleasure of working with Grant (one of the lead researchers on Merlin Sound ID) on other porjects in the past! The whole team
by gcr 3y ago
So funny to see this! I've had the pleasure of working with Grant (one of the lead researchers on Merlin Sound ID) on other porjects in the past! The whole team is kind-hearted and incredibly smart. Sound ID is one of the very few research projects that make me instantly smile hearing about in the news.
I've always been impressed with the responsibility that the Cornell Lab and Macaulay Library give toward involving people (citizen scientists, raters, and experts alike) in the curation of the data they use. This isn't an application where "hoovering up the entire Internet" is the right thing to do; it's a case where those hand-tweaked parameters make all the difference. A considerable amount of thought is given to every sound sample, and every person in the pipeline's working incredibly hard.
Hats off the team to publishing more about the internal process!
- woodson 3y agoThe article is very thin on actual technical details. It doesn’t describe how the spectrograms are processed or details of the neural network they’re using (other than using convolutions, which is pretty standard in CV/audio/speech processing). What is interesting to me, though, is that they mention the importance of more precise time labeling of sounds. Previous BirdCLEF Kaggle competitions just provided many-minute long recordings where all that is known (labelled) is that a certain list of species are to be heard, but not when in the recording. That’s not too useful as training data, as they point out in the article: This can lead to problems: if other species are singing in the same recording, the model will erroneously call all species in the recording a White-breasted Nuthatch, leading to false predictions.