The Sound of Air (part 1)

I am fascinated by the effect of air on sound. In particular, I’m interested in the effect it has on the sound of air and road traffic. It sounds like a ‘tumbling’ or ‘folding’ of sound, with fluctuations in loudness, direction, and filtered frequencies. Here is an example of the effect in a recording of F15e fighter aircraft dogfighting in Lakenheath, Norfolk, by Microscopia, available at The Freesound Project:

I hope to reproduce the effect by analysing it and understanding its components. This post represents my intial thoughts on the subject and some ideas on how it might work. The aim is to develop my understanding of the environmental acoustic phenomenon, and – if the effect can be modelled – use it as a tool for processing sound.

Let’s try to identify the components of the sound effect of air, beginning with some basic observations. We mainly find this effect in the sound of distant air and road traffic, but it also happens with other loud, broad-spectrum sounds heard at a distance, such as a massed crowd or fireworks, though the latter are so short that the effect is less apparent. Presumably it also affects quieter sounds, but those we cannot hear. It seems fair to assume that its intensity is porportional to distance; the farther away we hear a sound, the greater is the perceived effect. In this sense, we might say that it is a sonic equivalent to ‘aerial (or atmospheric) perspective’ – the blue-tinged haze that reduces contrast and colour in a distant skyline. Fog is a more intense version of the same thing. But a significant difference between visual and sonic perspective is that the latter is perceived to be in constant flux, and perhaps this is one of its fundamental characterstics.

In Soundscape (1994, p.134), R. Murray Schafer mentions the following environmental effects: reverberation, echo, “drift (fading)” and “displacement (ambiguous point of origin)”. These are all, I think, plausible characteristics of the sound effect in question. In response to my question about the effect, Audio Field Recorder highlights the complexity of the issue:

In the natural world outside there are far too many variables which cannot possibly be taken into account, all of which produce a significant effect on the sound waves as they travel outwards from their source, making it difficult to identify its true location. Some of these are variable air and ground temperatures and localized air pressures from wind effects along the path travelled by the sound waves, localized ground effects which affect wind speed, which in turn affects its air pressure; different frequency filtration by trees, shrubs, grasses crops, reverberation caused by natural ground contours, etc.

To model even one of these variables would be a formidable computational task, so the chance of being able to integrate them all into one effect seems very slim indeed. Nevertheless, it may be that it is not necessary to model all of these variables, because as yet we cannot say to what extent each variable contributes to the effect. To my ears, the most typical element is a sloshing kind of sound, a filtering of the low-mid frequencies that fluctuates up and down.  We may be able to visualise some of these fluctuations with a spectrogram of the recording of the F15 dogfight that we listened to earlier (image created with Sonic Visualiser):

Spectrogram of F15 recording. x: time, 1'30". y: frequency 20-20,000 Hz, log scale. z: amplitude, dBFS (0dB = white)

The spectrogram image represents the first minute and a half of the recording. Time is represented from left to right, the vertical axis shows frequency on a logarithmic scale, and amplitude is shown in colour – louder sounds are lighter. The bird sounds are easily spotted in the upper portion; the magpie calls can be seen to start around the centre of the image and at the far right a burst of its chatter is visible (click on the image to view it in greater detail). The long duration and slow fluctuation of the wavy bands suggests that they might be due to the sound source (the jet engine) rather than the effect of air at a distance. Notice the clusters of upward-pointing peaks that are more rapidly changing than those wavy bands. I think these visualize the most characteristic sound of the effect – the sloshing, folded, filtered sound. These visible forms do appear to take the form of filters since they appear as dark lines within other sounds. They appear more three-dimensional than the other shapes, perhaps because the peaks seem to occlude each other, like a mountain range. When the spectrogram is viewed in sync with the sound, these patterns appear to correspond to the sounds in question.

I cannot be certain that within these fuzzy images of peaks I have found the effect. Also, we don’t yet know what causes the effect, nor how we might model it. In the next part, I will analyse more sounds and see whether similar patterns are found in other recordings of loud sounds at a distance. It might be possible to use the same process that produces the spectrogram, but in reverse, to generate sound from an image of those clustered peaks.

C. Cannam and C. Landone and M. Sandler
This entry was posted in Audio, Auditory Perception, Nature, Research and tagged , , , , , , . Bookmark the permalink.

9 Responses to The Sound of Air (part 1)

  1. Audio Field Recorder says:

    A very interesting subject – however, the difficulty will be identifying those sounds affected by the wind, as it will have different velocities at different points in the soundfield.
    In the recorded soundfield above, the aircraft will all have been at different altitudes, bearings and distances to each other and to the birds and water feature, so would all be affected differently. The shading (occlusion) of frequencies referred to in the spectrogram could be due to the fact that similar recorded frequencies are superimposed from different locations, some of which have been affected and others unaffected.
    One way of limiting this problem would be to analyse narrow soundfield recordings made with parabolic dish microphones.
    Looking forward to the next part!

  2. anonymous says:

    swept comb-filter, known more formally as phasing.

    “Commonly, flanging is referred to as having a “jet plane-like” characteristic”

    this answer seems too obvious to be what you’re looking for, so i assume you already know these terms and effects, i don’t really get what more you want to know…?

    • Guy says:

      It’s more a case of “too simple” than “too obvious”. What I’m trying to understand and recreate is how sound changes as it travels long distances through the air. If you take a recording (e.g. the sound of a jet engine at close range) and process it with either flanging or phasing, then it doesn’t give the impression that the sound is a long distance away. Instead, it just sounds like phasing or flanging (I know because I have tried it). Spectral analysis of distant sounds suggests that some kind of phasing/flanging does happen in these situations, but those effects alone do not fully account for the phenomenon in question. Some other processes must be happening to the sound as it travels, so I’d like to determine which ones. Also, it is not entirely clear why flanging or phasing occurs in these situations (it could be to do with destructive interference with sonic reflections, for example, an/or a result of the varying speed of sound in air via temperature and pressure fluctuations). If it is possible to understand what is happening and why it is happening, then it might be possible to recreate this effect.

      • Sonic Fields says:

        Hi Guy. Your comment ‘…it doesn’t give the impression that the sound is a long distance away..’ has made me think a bit more about what is happening.
        All the known sound propagation-related factors such as the inverse square law of pressure decay v distance travelled, airflow speed v pressure variation, reflection and refraction, together with its directional determination through inter-aural time delay and amplitude difference, does not assist in determining the distance between the receiver and sound source. I believe we are only able to determine an approximation of the distance to the sound source [like the jet, or the buzz of a fly] only because our memory is able to recall previous experiences we’ve had of that particular sound with which comparisons can be made.
        If it’s difficult making a previous memory-based comparison, then it will be difficult judging how far away the sound source is. eg: Take the rumble of thunder – it’s very difficult judging how far away the storm or source may be, as memories of previous experiences would still have no distance association made with it as it is a rather indeterminate source unlike the clap of thunder from a visible lightning strike. Yet we know how far away the high-pitched whine of a gnat is from our ear – it’s close – we are ready to take a swipe at it – as we know it’s going to bite – all from previous experience.
        Taking it on a step further – shell shock from WW1 & WW2 – the scream off the shells and bombs caused psychological shock as previous memory-based experience of that particular sound meant that something terrible was about to happen – the distance away from the sound source [shell or bomb] could not be precisely determined enough for the hearer to know that it would land far enough away not to cause death or injury.
        Just my thoughts!

  3. Guy says:

    Thanks for your thoughts, Lawrence – they are really quite useful. I also enjoyed reading and listening to your recent post on ‘Flight’. It seems to me that your all-inclusive approach to field recording is similar to that of the Pre-Raphaelite painters, who grouped together in opposition to the subject-centred approach to painting, opting instead to include all the background details of a scene.

    The idea that perceived proximity is dependent only on memory is certainly plausible, but I feel that I need to do more research on the subject yet. Furthermore, there are some effects (e.g. Doppler) that may provide auditory clues to sound source proximity. But then again, it may be that these cues are also dependent upon remembered comparisons. If your suggestion is true, then it implies that my project of hoping to recreate the sound effect of air is doomed to failure, since it rests on the assumption that the effect is objective (to some degree, at least) and not wholly subjective, as you suggest.
    [edit] Actually, it’s probably not so bad as I thought, even if the effect is mainly subjective, because we all interpret the same sounds in approximately the same way. Food for thought, and lots more to explore!

    Part 2 of this exploration is long overdue, and so I shall endeavour to pull it together and post the results of my analysis and initial experiments as soon as possible.

  4. Mike says:

    The simplest solution for this would be to use a random (smooth/filtered noise random, not S&H) LFO to automate the various parameters of a phaser, flanger, and a notch filter (band reject), and have each of them sweep through the spectrum, at different rates, and automate the mix of each so they fade in and out randomly as well. You could pretty easily do this in Max/MSP or pd. If you use Ableton Live Suite, it has Max for Live built in, so you could create an effect plugin like this to use on whatever audio you put on a track.

    • Guy says:

      Thanks for the suggestion, Mike. I’ve not picked up this thread for a while, but might give it another go now. I have actually tried something of that sort, with some success: I used automated, randomized control of noise sources, filters and banks of flangers in Audiomulch. Whilst it does generate the churning comb-filtered effects quite well, it’s a bit more difficult to create a sense of movement and place, i.e. where the sound is coming from and where it’s being heard at. I suspect that in real-life conditions, the relevant parameters are to some extent interacting, and not fully independent of each other, which is the case with purely randomized control. There’s more exploring to be done. Cheers!

      • Mike Max says:

        Ah yes. Spatial placement. You could try ending the chain with an HRTF panner like Wave Arts Panorama, which will allow you to automate the movement of the sound through the stereo field (not just XY like standard LR, or even 5.1 or 7.1 pan, but also Z, or height). Maybe also add an outdoor reverb at the end of that using Altiverb or Speakerphone to give it that subtle but detectable presence in space. The other parameters sweeping the spectrum are somewhat random due to phase cancellations caused by environmental reflections. But the 3D pan should be a smooth sweep, and reverb settings should be static at the end of the process. That’s probably as close as you’ll get without actually creating a software simulation of an outdoor space that will calculate ray-cast RT60 for densities and materials of surfaces on the fly, as well as doing the math for changes in air density and atmospheric pressure between the listener position and the source to be applied to the source as a filter. Of course at that point, you’d need a really powerful computer if you wanted it to run in real time, or without significant latency.

      • Guy says:

        I’ve looked into HRTF a bit – computational and price costs were prohibitive at the time. Good idea to try some reverb too – I’ve tinkered with convolution reverb, using a woodland IR, but for this application you’d ideally need an IR of a big sound over a big area … oh, like a thunderstorm – could use a thunder crack as an IR. I’m going to try that now!
        As you say, software sim would be ideal. I guess that kind of approach is what I’m aiming at really – modelling the process rather than just recreating the sound.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.