I worked on a bit of a side project to help some local reseachers this weekend: Getting a 10 X speed up on an important audio analysis tool for endangered species conservation. The Northern Spotted Owl is endangered. Traditional monitoring methods—callback surveys and mark-recapture—stress the owls and are becoming less effective as populations decline.

The USDA Forest Service now monitors 4,000+ sites annually using passive acoustic monitoring instead. In 2023 alone, that generated 2.2 million hours of audio. Processing that data is a bottleneck. PNW-Cnet is the convolutional neural network that classifies these recordings, identifying spotted owls, barred owls, and 80+ other species.

I converted it to ONNX format and documented the process. https://github.com/breadboardfoundry/pnw-cnet-5-onnx

Maybe it will help if you are tinkering in machine learning or audio processing.