Generate a continuous sound of car rolling noise from a recording
What would be your approach of generating a continuous sound of the car using an audio recording made at a fixed distance from the road, during a car pass-by with a constant speed?
The end result would be a car rolling noise as if it was recorded outside of the car, that is the microphone travelling with the car. Either an infinitely long synthesis or a looped sample in which one can not recognise the restart point.
The audio recording that can be obtained has an increasing level to a point where the car is immediately in front of the observation point, the microphone, after which the level decreases to 0. We need to achieve a constant level, potentially based on the fft analysis and resynthesis of the short sample when the car is in front.
I would place an omni mic in the middle of a roundabout, and drive the car around it. Capture a few circuits, find the best bit, and loop it. If you want a constant rolling sound of a car, that would be a practical way to achieve it.
That is an interesting idea, but you can only have a certain limited speed on a roundabout. The goal is to reproduce a car pass-by at a motorway speed using level change and panning, or spatial localization depending on how far is the observation point. So it would be important to have a car traveling at a speed where rolling noise has a dominant contribution over engine noise. I am also interested to apply a speed correction which is the change in the frequency content. The faster the car goes the higher frequencies are generated.
i´ve done it like that for a jet fighter: one sound from the front, one from 90 degree, and one from back.
then when it passes by i correctly calculate the angle and mix the 3 audio channels likewise.
Would you have to correct for the doppler shift?
Antonio, Doppler shift may indeed be an important consideration for a fast moving object approaching listener, the observation point. I imagine such an effect would be a frequency shift implemented in FFT analysis. However, for broadband sound such as a car rolling noise that has no distinct frequency components the Doppler effect may not be noticable. Subjective comparison of the generated sound with a recording would give an answer, which I am happy to report once I complete my task.
But I am still unsure how would generate the constant car rolling noise based on a short recorded sample. What Roman did with his fighter jet may be relevant.
Would a convolution of the white noise with the short recorded sample of car rolling noise give me the expected output ?
dont underestimate the doppler effect. and the air absorbtion.
and for the sound source use 2 channels for the right and left ear and apply all the geometric calculations to them independently. no matter if for mono, stereo, prologic, atmos or ambisonics - phase, pitch, loudness are fundamental for movement.
Roman, I think I am still one step behind you. Before I apply geometric calculations, how do I generate a constant sound of car rolling noise based only on a 2 or 3s long sample when the car was in front of a microphone 15m from the road??
how do I generate a constant sound of car rolling noise
it's a complex task... i would divide the sound into smaller parts - simplest example:
the continuous rolling of the car is a steady kind of noise (from engine, wind-rush etc.) mainly in the middle to low end
while the contact between the tires over any gravel or other irregular parts of the surface it drives on causes more granular-like sounds in the higher ranges of frequency
perhaps you can sample the recording and create the middle/low end by an FFT freeze(modulated with a bit of noise within the FFT to make it not so regular - see the phase-vocoder tutorial on this site), but then also use the recording in granularly-sampled sounds for the high-end; FFT is not as detailed in its analysis for high-end resynthesis as in the low-end, so my recommendation would be to combine several techniques to both analyze and recreate the sound, once you have a steady sound of a car rolling from these different parts, you would then employ Roman's math suggestions to shape the listener's perspective of everything.
if it was recorded from 15 meters away, it will - in theory - be impossible to substract the air absorbtion if you want it to be closer.
but farther away is always the same, and is probably easier than you might think.
the "geometry" for amplitude, filtering and doppler is also always the same. think of the user as the center point of a circle (or half a circle for stereo), then the distance to the source is always the radius.
let´s ignore the frequency filtering for now and only look at the amplitude: amplitude is always relative and exponential, so in db it is a linear fade if the source moves away with a constant velocity.
if you know that a recording of the source was from 15 meters away, and you move the virtual source 30 meters away the gain it had to be lowered by 6 db. if it should come closer for 50% - 7,5 meters - you would add 6 db.
regarding the freqeuncy filtering from the air absorbtion when coming closer than the original was brings a little issue with it, at least i would not have an example ready how to do that. (eventually you could just introduce a max distance of 15 meters in your software.)
if you have the chance to do that, try to catch sources from 4 or 5 meters distance (if it is a car, a flute could be recorded from 1 meter) and make sure it happens in a free field situation and not in a garage which results in 200% reverb. ;)
my personal favorite trick when placing virtual sound sources in a soundfield (i.e. for stereo or more) is to add 2-3 simple, not so loud reflections, as if there would be small wall sections somewhere in the area, and process those reflections together with the moving source to help the brain track the direction. it doesn have to be correct phases or other special treatment, just any sort of echo with the right delay time.
Andy Farnell's book Designing Sound https://mitpress.mit.edu/books/designing-sound may be useful. Especially the section that walks through procedural audio generation for a helicopter flyover. The book uses Pd, but there are Max ports of the patches available
+1 to the Farnell book, I just got it recently and it's awesome!