Question about post-processing: quiet birds suddenly screaming (1 Viewer)

[OkamotoKeitaSin]

Sometimes when recordings birds - usually when targetting the (presumed) contact call of a foraging flock - one or two individuals would have sudden outbursts of loud songs/calls, only to return to their contact call again after a few bouts.

This causes problems when normalising the sound recording during post-processing. The loud song/calls will set the benchmark, making the rest of the sound clip to be soft and barely audible when it wouldn't have been an issue otherwise.

What's the recommended method to post-process such sound files?

Should I:
Split these recordings into multiple clips and normalise each of them to -3dB?
Deal with it and just normalise the entire recording?
Use some function to reduce the amplitude of the loud call/songs?

 

[Jon.Bryant]

If you have a bird transitioning from a quite call to a loud song, I would personally;

1/ Keep the recording as is and normalise the whole.
2/ Create shorter clips of the call and song elements and normalise these separately.

With this you have a recording which has context about the volume of the call compared to the song, and other (shorter) recordings which focus on the call or song only.

I would not try to keep a single recording but to tone down the song prior to normalisation. This would not be so easy to do seamlessly, and by doing this you will create the impression that the bird either has a very loud call or very quite song, which would not be representative of the bird’s true vocals.

 

[Mono]

If you are just wanting the contact call then just edit out the unwanted bits. If you want a representative recording of a flock in flight then, as Jon has said above, you need to adjust the levels to keep the level differences between the different calls.

 

[OkamotoKeitaSin]

Thanks for the advice Jon and Mono. Hmm I see, quite context dependent as shared by the both of you. But okay it does make sense that I shouldn't artificially reduce/increase the amplitude of respective sections that would distort different vocalisations' respective volumes!

What's the general consensus regarding uploading so to speak "identical" recordings online? If I do create two versions - one with the whole thing normalised, one with softer call only, one with louder song only - there'd be a total of three files of the same incident. Would it make sense to upload all three of them? (Asking because I'm completely clueless about concerns regarding server space/other issues I might not have even realised).

 

[Jon.Bryant]

If you upload to Xeno Canto you can add comments, so you could cross reference an already uploaded recording. I have never heard that server space is an issue, so wouldn’t worry about that.

I think what you upload, really depends on quality and/or rarity. If you end up with two or three good recordings, or poorer recordings of something not well represented in the collection, I would submit them all. On the other hand I don’t submit everything I record - I don’t really think collections are meant to be repositories for my poor recordings of common sounds! I fill my own hard disks full with a lot of stuff. Probably a bit contentious, but I think far too many people are uploading poor quality recordings to Xeno Canto, often from NocMig sessions. I am not convinced this really adds anything to the collections.

 

[steinn]

Using 32-bit flow able recorders you can reduce the level nicely on the loud bits and level the low bits of the recording. A bit difficult to make it sound natural since u also raise the natural noise/background sounds. It works great in Audition to take bits of the recording to wanted level
Stein

 

[OkamotoKeitaSin]

Hi Stein, when you say levelling the loud and low bits, does that mean normalising different parts of the recordings to -3dB (or whatever value) respectively? Am a little puzzled because wouldn't that, as you shared, cause the background noise to fluctuate accordingly?
Also wondering why that's only possible on 32 bit float but not on 24 bit.

 

[Jon.Bryant]

I think you are right that 24bit would have enough head room to increase the gain of the quiet bits, but also right that this would also create a jump in the background noise, so would not sound very good or natural.

With 16 bit the noise floor is -96dB and with 24bit -144dB, meaning that you can’t record anything quieter than -144dBFs. With 32bit float the noise floor is a lot lower. If you made a mess of the recording levels with 24bit, the quiet sounds could potentially disappear below the noise floor (but this would need to be a really gross error) - when you normalised the recording you would then be amplifying the noise floor rather than the lost quiet sound. With 32bit float it is virtually impossible for this to happen - this plus the virtual impossibility for the signal to be clipped, is why recording levels become irrelevant. But if you have a recording with a mx of loud and quite bird noises, then recording level was obviously not grossly wrong. I then think background ambient noise (even in a very quiet environment) will be well above the 24 bit (or even 16 bit) noise floor, so there would be no advantage of 32bit float when it comes to adding gain in post - the ambient noise will drown out any extremely quiet sounds that may have been lost below the noise floor.

I think the same type of logic applies to the argument that you need ultra quiet mics to make the most of 32bit float. Nearly all mics have noise levels below the ambient noise of a quiet environment, so with field recordings the ambient noise rather than mic noise or the noise floors is the real problem.

Regards

Jon Bryant