Understanding audio speed modification

[jezharringtonhayes]

Hi there. Firstly, thank you for this great open-access software. In general, I'm finding it easy to use and incredibly useful for the audio monitoring project I'm working on, so apologies if the following sounds a bit negative: I have what is probably a very basic question, but I'd be surprised if I were the only beginner struggling with it.

1. The audio speed modification scale runs from -10 to +10, with +1 being the default speed, and apparently "negative values slow down, positive values speed up" (should the default setting therefore not be 0?). Reading through the official documentation, I've discovered that -10 means one tenth (0.1) of normal speed. Am I right, then, in deducing that -5 would be 0.2 x normal speed, -2 would be half normal speed, and anything between -1 and +1 (including 0) is essentially the same as +1? If so, I find this quite unintuitive, and wonder whether a scale between +0.1 and +10 (with +1 remaining the default) might be easier to understand for people new to the software.
2. Much more importantly, I'm struggling to understand how modifying the audio speed affects the min. and max. bandpass frequency. I have audio files recorded at a high sampling rate (32 kHz - as recommended for diurnal forest birds (which I'm studying) by NZ's Department of Conservation, the manufacturers of the AR4 device I'm using). As the BirdNET-Analyzer allows a maximum max. bandpass of 15 kHz (which, for all of the below, I kept at 15 kHz, and min. bandpass at 0 kHz), I assumed that in order for the software to additionally consider the range of 15 - 30 kHz in my recordings, I would simply need to halve the audio speed (which, if my first point above was correct, would mean the -2 setting). My intuition inferred that slowing down the audio speed increases the frequency range which can be analysed by the software (as in, a true range of 0-30 kHz would come out as 0-15 kHz at half speed) and that speeding it up reduces the frequency range, but is a nice way of shaving valuable seconds off the processing times for people lucky enough to be analysing whale vocalisations. Clearly, I'm wrong. In the file I'm using to experiment with various parameters, at speeds -1, 0 and 1, two bird species are identified (silvereye (Zosterops lateralis) and Eurasian blackbird (Turdus merula)), however at -2 and below (as well as +2 and above) no birds are identified at all, and, interestingly, at -1.1, one bird (which I know to be a Eurasian blackbird, albeit not the same 3-seccond segment identified at speeds -1, 0 and 1) is identified as a New Zealand kākā (Nestor meridionalis), which, as a quick sample of xeno-canto recordings confirms, vocalises at lower frequencies than the Eurasian blackbird. So my theory that reducing the audio speed increases the maximum detectable frequency is evidently wrong, and I have reason to believe that when modifying the audio speed, BirdNET-Analyzer does not "account for" the modification, and takes the distorted frequency at face value (as in, what would be a blackbird at normal speed becomes a kākā at marginally slower speeds). Which begs the question: is there any way to have the software analyse recordings in the range above 15 kHz? I would be hugely appreciative of any information you can share on this topic. Thank you.

[MacJudge]

I'd like to comment on your second question. If you have recordings at a sampling rate of 32 kHz, than you cannot safely reconstruct audio signals at a frequency of 16 kHz or above, so the bandpass filtering at 15 kHz will be almost no restriction for you. See Nyquist–Shannon sampling theorem: https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem

[jezharringtonhayes]

That makes sense now. Thank you so much!