Best practice for measuring amplitude of biophony

Question

It is often useful to quantitatively measure (or at least estimate) the amplitude of an acoustic event such as a bird song. The desired measurement could be, for instance, that a bird sang at 105 dB SPL (re 20 uPa) at 1 m.

However, I am unsure how to best estimate the sound level of an event, even if it is possible to record the event at a known distance with a calibrated microphone or a sound level meter. For the purpose of this question, let's assume that we record a sound event originating from a point source which has well defined boundaries in frequency and time and a high signal to noise ratio above background noise. We record the event exactly 1 m from the source. We are not concerned with directionality, we assume all measurements are “on axis”. How should we quantify the sound pressure level in dB SPL @1m (or some other measure of sound pressure) of the event?

Specifically,

1. Should we measure some average (such as rms) over the duration of an event or the peak value?
2. Should we take a measurement at specific frequencies or across the frequency spectrum?
3. Should the background noise measurement be subtracted in some way from the signal measurement, as in the methods of [1]

[1] https://besjournals.onlinelibrary.wiley.com/doi/pdf/10.1111/j.0021-8790.2004.00814.x

Answer 1

I am afraid the general answer to your question is: it depends.

The issue of choosing a suitable metric to describe the intensity of an acoustic phenomenon is quite common in the field of environmental noise. Let's take the example of road traffic noise. When dealing with a continuous flow of vehicles like on a highway, it is quite common to use either equivalent sound levels with time constants that can reach 24h, or fractile sound levels. When dealing with isolated vehicle pass-bys like in the evaluation of the performance of road pavements, the standard way is to evaluate the maximum sound pressure level - that should not be confused with the peak level - or the Sound Exposure Level (SEL). In general the metrics used will vary from one transportation mode to the next, not to mention industry noise. So it depends on the type of signal, but it may also depend on what feature you want to highlight.

In the field of environmental noise on land, the use of the peak level is deprecated because the reproducibility of the estimation of peak level is considered lower, owing to the fact that peak level is based on a very short fraction of the signal. Moreover, the peak you will register will be sensitive to the equipment you use. The impulse response of the equipment might play a decisive role here. So for short events a possibility is to compute the maximum sound pressure level as defined in IEC 61672-1.

But whatever the metric used, it will almost always involve an integration over a time interval (this applies to the evaluation of the rms value) or depend on a time constant (like for the maximum sound pressure level). When reporting a dB value, it is essential to specify this parameter (for instance by writing Leq,10s=XX dB (re p0) for a 10-second equivalent sound level). Otherwise the dB value reported will be of limited use, because the dB value will depend on the time constant in general.

Subtracting the background noise might be considered if the SNR is such that the background noise influences the total sound level. But then you really need to know what you are doing.

Answer 2

The first two areas of questions can be answered like this:

Describe the sound event in such a way that the information provided can support the detection and classification of the event/signal and the species. The amount of features to be described will depend on the complexity of the signal.

However, I would suggest, to always consider:

• temporal features: peak level, duration, peak location in signal, RMS level
• spectral features: min-mean-max frequencies, with level and locations in signal, number of loops.

Edit: following reformulation of OP I would augment my answer: Any description of bird/animal calls should describe quantity and quality:

1. If call is short (large amplitude variation) provide peak-level; if call is long with or nearly constant amplitude, provide RMS value; provide both if in doubt
2. describe the spectral functionality in sufficient detail, but do not try to provide spectral time estimates for broadband sound due to reciprocal uncertainty (uncertainty principle)
3. do not subtract noise from signal. BTW this is NOT done in cited paper, where the authors correctly measured the background noise and considered it as co-variants, but they did not subtract it from signal.