Hello!

I could be wrong, but from looking at the source code of the ACI calculation on the scikit-maad it appears that it autosets j = length of signal. If you want to break it down to e.g. 5 windows you can break the spectrogram into 5 chunks along time and then sum the ACI in each window. This gives a more similar result to the results you got using the other methods. What did you set as nbwindows for seewave?

s, fs = maad.sound.load('test_data_ACI/LINE_2003-10-30_20_00_34.wav')

Sxx, tn, fn, ext = maad.sound.spectrogram (s, fs, mode='amplitude')

full_time = Sxx.shape[1] # number of time samples in spectrogram

j_count = 5 # number of chunks we want to break it into

window_size = np.floor(full_time/j_count) # total time divided by number of chunks = number of time samples per chunk

ACI_tot = 0

for i in range(j_count):

_, _ , ACI = maad.features.acoustic_complexity_index(Sxx[:,int(i*window_size):int(i*window_size+window_size)])

ACI_tot = ACI_tot + int(ACI)

This gives ACI_tot = 1516

Quick question on this topic to take advantage of those that know a lot about it already. So once you have extracted all your camera data and are going through the AI object detection phase which identifies the animal types. What file formation that contains all of the time + location + labels in the photos data do the most people consider the most useful ? I'm imagining that it's some format that is used by the most expressive visualization software around I suppose. Is this correct ?

A quick look at the trapper format suggested to me that it's meta data from the camera traps and thus perform the AI matching phase. But it was a quick look, maybe it's something else ? Is the trapper format also for holding the labelled results ? (I might actually the asking the same question as the person that started this thread but in different words).

Another question. Right now pretty  much all camera traps trigger on either PIR sensors or small AI models. Small AI models would tend to have a limitation that they would only accurately detect animal types and recognise them at close distances where the animal is very large and I have question marks as to whether small models even in these circumstances are not going to make a lot of classification errors (I expect that they do and they are simply sorted out back at the office so to speak). PIR sensors would typically only see animals within say 6m - 10m distance. Maybe an elephant could be detected a bit further. Small animals only even closer.

But what about when camera traps can reliably see and recognise objects across a whole field, perhaps hundreds of meters?

Then in principle you don't have to deploy as many traps for a start. But I would expect you would need a different approach to how you want to report this and then visualize it as the co-ordinates of the trap itself is not going to give you much  information. We would be in a situation to potentially have much more accurate and rich biodiversity information.

Maybe it's even possible to determine to a greater degree of accuracy where several different animals from the same camera trap image are spatially located, by knowing the 3D layout of what the camera can see and the location and size of the animal.

I expect that current camera trap data formats may fall short of being able to express that information in a sufficiently useful way, considering the in principle more information available and it could be multiple co-ordinates per species for each image that needs to be registered.

I'm likely going to be confronted with this soon as the systems I build use state of the art large number of parameter models that can see species types over much greater distances. I showed in a recent discussion here, detection of a polar bear at a distance between 130-150m.

Right now I would say it's an unknown as to how much more information about species we will be able to gather with this approach as the images were not being triggered in this manner till now. Maybe it's far greater than we would expect ? We have no idea right now.

but no "time scaling" adjustments to adjust a too "slow" or too "fast" video ?

Thank you so much for looking into this issue quickly! Much appreciated.