I recently had a few ideas about the frog project. I have an issue with the sound: the ambient sounds of nature are usually very low. This has two consequences.
First, the recorded audio of nature sounds is so low that it becomes really noisy, such that we don't really recognize the sound anymore. To put it clearly: the sound is bad...
Second, it's hard to trigger the sounds according to their loudness because there is a huge difference in decibels between the sounds of nature (very low) and the sounds emitted by the devices themselves (very loud). If we want to collect sounds, we can't really use an audio threshold (ie. pick only loud sounds) cause then, we'll end up with sounds emitted by the devices themselves only in the collections. They will take all the storage space, leaving none for anything new.
The first issue could be solved by trying to reproduce the sound using waves. Waves are clearer and louder than recorded audio. We could set the waves frequencies using FFTs. I have already successfully FFTs on AVRs when I was in residency in Quebec City (see this link for example). We could simply generate a wave with a frequency corresponding to the peak of the spectrum (maybe smoothed with its surrounding bands).
The second issue could be fixed by treating very loud sounds (ie. the ones emitted by speakers) differently than other sounds. Those sounds would be stored only temporarily (in a buffer) then repeated just once.
The issue remains on how to actually choose the sound samples. The following could be used as guidelines: