5
$\begingroup$

For click classification in PAMGuard I've used a trigger filter from 12Khz to 64 khz. In the click classifier I've used as the test band 15000 to 64000 and as control band 0 to 10000. I wanted to know if the fact of using a trigger filter from 12000 to 64000 would delete all frequencies components lower than 12000. If this is the case, it wouldn't make any sense to select as control band the range 0 to 10000 as there won't be any signal there.

$\endgroup$

2 Answers 2

2
$\begingroup$

The trigger filter does not eliminate sounds outside of the range that the user has set it to. The trigger filter just tells your detector/classifier where to look for signals. In other words, your trigger filter of 12-64 kHz does not act as a band-pass filter (this is the job of the "digital pre-filter" in the click detection phase in PAMGuard, which does filter your wav snippits according to whatever filter you've specified).

The control band of the trigger filter is used to compare energy in the trigger filter's search band. For example, if there's energy in the 12-64 kHz band, but also energy in the control band (which you've here set to <10 kHz), then a click is less likely to be classified as species X. If, on the other hand, there's energy in your 12-64 kHz trigger band and there's not much energy in your control band, then the click is more likely to be classified as species X.

$\endgroup$
0
$\begingroup$

I tried without success to see something obvious in the PAMGuard code, so I suggest that a trigger filter 12 - 64 kHz (looks like beaked whales) should not be considered as preprocessing that eliminates all other frequencies.

If there is sufficient signal in the 12-64 kHz band, a click will be called for, and a (beaked whale ?) will be classified if there is no significant sound below 10 kHz. If the trigger filter will remove outside band energy, no dual-band classification would be possible.

In case of in-band classification, it would make no sense to call the second band control band, and you would need a larger feature vector.

Edit: This answer is consistent to the one by @Chloe

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.