What is comb filtering? What does it sound like?

Comb filtering happens when a sound mixes with its own reflection, or a signal is mixed with a delayed version of itself. It can sound subtly bad; it can sound absolutely dreadful. The one thing it never sounds is good. If you watch the video and hear the demos then you'll be prepared for the worst.

Automated transcript

Learn audio online with audio masterclass audiomasterclass.com comb filtering can happen when a sound mixes with its own reflection or when a signal mixes with a delayed version of itself in this video i'm going to demonstrate what comb filtering looks like and what it sounds like my sound source to start off with is going to be a frequency sweep a sine wave going from 20 hertz to 20 kilohertz let's hear it

i heard some rather interesting distortion at the end maybe it's my system did you hear it if you did hear it leave a comment and tell me what you heard i might investigate it in a future video what i'll do now is create an auxiliary track and insert a plugin so we can have a look at what's going on as well as listen i'm going to route my sine wave to the auxiliary track and just test it to make sure okay there we have it and we've got the sine wave in the oscilloscope section of this plugin i'm going to use what's called spectrum here what i would normally call it a spectrograph and a spectrograph is a device or software that produces a spectrogram so let's have a look at the spectrogram and see what we get

okay that's far enough for now it's in the nature of spectrographs that there are some errors in what you see so first we'll check the level over here we can see that it's coming in at -12.0 db it means dbfs

so the level is consistent all the way through or at least as far as we need to go but if we look at it here what we can see is this peak level it seems to diminish as the frequency gets higher

okay so that's just an error in the spectrograph so we won't bother about that we'll just expect and accept it as it is for the moment so what i'm going to do now is set things up so that this sine wave sweep mixes with a delayed version of itself so what i'll do is duplicate the track and i'll zoom in and i'm going to delay the copy by five milliseconds

okay so now we can listen again and let's see what we get

okay that's far enough so this is comb filtering and it's called comb filtering because the chart looks like the teeth of a comb so at some points in the frequency scale the two sine waves are in phase and they add up together and double which means they rise by 6 db and at other points in the frequency scale they're 180 degrees out of phase so they cancel out pretty much completely let's have a look at a point where they combine constructively so i'll go to 100hz here we are at 100hz and i know it's here because i checked earlier so if i zoom in

we can see there

we can see there that we've got this peak on the top and this trough on the bottom and they're going to cancel each other out let me just zoom in a little bit on the height of that so you can see it better there we go so we've got the peak on the top and the trough on the bottom which will cancel out now i'll go to 200 hertz so we'll see what happens there so once again this is where 200 hertz is in the sweep and i checked that earlier to save time in the video so i'll magnify that and magnify it again and give it a bit of height okay

okay there we are so we're not quite in the right place there but we're near enough to see that here we've got a peak and here we've got a peak just slightly out of time there and they will add together causing an increase in level so this is what we get we get at 100 hertz 300 hertz 500 hertz we get the dips in level and they look like the teeth of the comb and at 200 hertz 400 hertz 600 hertz etc that's where we get the peaks clearly this will change if the delay time changes and that's something you might want to try for yourself so let's listen to it and watch it again

she's

i love that distortion at the end i think i'd like to try and do something creative with it one day okay so i'm going to try a similar test and this time i'm going to do it with white noise so let me set that up so here's one track of white noise so it's standing in for a music signal or speech or whatever you might be recording in real life it sounds like this

okay we don't need to hear any more than that now i'm going to duplicate it and delay it i'm going to delay it by 5 milliseconds that's five milliseconds it doesn't matter what the exact delay is and i would certainly encourage you to try it for yourself and let me know in the comments how you get on so before i play it just have a little think what's it gonna sound like and what's it going to look like okay here we go

so it's a bit of a racket but i'm going to talk over it and mute it a little bit maybe so you can hear me clearly so here we can see we've got a dip at 100 hertz thereabouts and at 300 500 700 900 exactly as before and we've got these peaks at 200 400 600 800 etc you get the idea it sounds pretty dreadful doesn't it let me just compare white noise by itself with the comb filtered version so i'll use the solar buttons here so that'll be just the noise by itself and that'll be the comb filtered version here we go

so if it's going to do that and make that horrible sound just with white noise what is it going to do with music or speech we'll find out in a moment so let's just try adjusting the delay time and see what different sounds we get from this firstly just for completeness i'll make it so there's zero delay what you'll hear will just be louder

okay so let's try one millisecond of delay

that is just so different two milliseconds

i don't like that i don't like that at all three milliseconds

they all sound bad it's a question of taste as to which sounds worse

let's go up to a higher delay setting let's go for uh where are we

10 milliseconds see what that sounds

like it's getting a bit of a buzz effect let's try 20. there's something going on really deep down in the base let's try let's try more let's go up to 50 milliseconds see what we get 50 milliseconds i mean it doesn't have to be accepted there we go

i think what we can hear from that is that it's those short delay times that are the real killers and the longer delay doesn't seem to matter so much not that it doesn't matter at all it just doesn't seem to matter so much let's try a longer delay so where are we there 103 doesn't have to be exact

okay with white noise it sounds pretty much the same at this point but with a normal audio signal whether it's music or speech you would start to hear the delay as an echo which would be another bad thing there's so many bad things that you need to avoid in audio but that's part of the fun isn't it so what i'm going to do now is i'm going to use a speech signal and see what it does see what comb filtering does to speech let's see if i can find a poem so here we have a poem it's me recording in abby road studio two and i can't remember off the top of my head which microphone it was but if you'd like me to find out leave a comment and i'll dig deep into the archives and find out which microphone it was let's listen to the poem just by itself i wandered lonely as a cloud that floats on high or dale and hill when all at once i saw a crowd a host of golden daffodils beside the lake beneath the trees fluttering and dancing in the breeze okay yes i know i got the words wrong i like to tell people that i was reading from wordsworth's first draft of the poem so that's my excuse i'll just play it again because there's something we need to look out for i wandered lonely as a clown so if you look here you will see peaks and troughs in the response a crowd that's nothing to do with comb filtering this time because i was well away from any reflective surfaces that's just the harmonics in my voice so where are you seeing peaks that's where the harmonics is strong and where it's troughs that's where they're weaker i wondered lonely as a cloud so therefore floats on high or dale and hill when all at once i saw a crowd a host of golden so that's not the comb filter we don't need to worry about that so i'll duplicate the track just like i did before and once again i'll delay it and i'll go for the five milliseconds again so let's take those solos off

so let's listen to it now and see what we get in the spectrograph display i wandered lonely as a club that's dreadful that's oh that's dreadful it's just so dreadful but we need to listen to it i wandered lonely as a cloud that floats on high are dale and hill when all at once i saw a crowd a host of golden daffodils beside the lake beneath the trees fluttering and dancing in the breeze okay that sounds bad so as before it's probably a good idea to try some different delay times so without any delay at all i'll do the the solo thing again so without any delay at all it's going to sound like this i wandered lonely as a cloud so it's just six decibels louder that's all so if i now delay it by one millisecond so when that solo is active you're hearing just the original when i switched that off here in the comb filtered version i wandered lonely as a cloud that floats on high or dale and hill when all at once i saw a crowd okay it's bad try this i wandered lonely as a cloud that floats on high or dale and hill when all at once i saw a crowd a host dreadful let's try five milliseconds i wanted lonely as a cloud that floats on high or dale and hill when all at once i saw a crowd a host of gold okay just as i did before i think it's worthwhile trying the longer delay time so let's try so let's try 20 milliseconds so that's quite a long delay that would correspond in sound traveling in air to a path difference of 20 feet or 20 times 30 centimeters if you can work that out in your head six six meters six meters approximately so let's see what that sounds like i wandered lonely as a cloud that floats on high or dale and hill you see what we're getting now is an echo so it's not changing the the texture of the sound so much is adding the echo but we'll listen again i want it lonely as a cloud that floats on high or dale and hill when all at once i saw a crowd a host of golden daffodil so if we go further to 40 milliseconds now let's go to 100 milliseconds

okay we spot on a hundred there i wandered lonely as a cloud so at this point it becomes an echo and the comb filtering has effectively vanished so i think we've been through pretty much all of it we've tried sine waves we've tried white noise and we've tried speech if you'd like me to do a demonstration on music then i can make another video and add that but it will probably sound pretty much like we've heard it already comb filtering as you've heard is not a good thing and should be avoided where calm filtering typically would happen would be when you're recording if you've got a hard surface nearby then the reflection from the hard surface will mix with the original signal after a delay and that could happen if you if you're speaking at a desk and the microphone is mounted at pretty much the normal distance from your mouth then the mic will pick up the direct sound from your mouth and it'll also pick up the reflection that bounces off the desk after a short delay and the path difference could be something in the region of maybe i don't know 30 centimeters something like that which is one foot and sound travels at roughly one foot per millisecond so you've got a millisecond of delay there the other place you'll get comb filtering typically let's say you record an instrument in stereo with spaced microphones but the microphones are slightly different distances from each other then there'll be a delay between the microphones if you listen in stereo what that'll mean is that the instrument seems to be panned either to the left or to the right because of that time difference and the slight level difference as well but if there's some reason why you have to collapse that stereo into mono suddenly you've got that signal mixed in with a delayed version of itself and once again you'll get comb filtering just to round off i'm going to do something which is a little bit more realistic i'm going to try and mimic the situation where you're recording at a desk so i'll use the same poem which wasn't at a desk but i'll treat the delay so that it is similar to the way you might expect an echo from a desktop surface to sound so i'll go back to one millisecond so that's replicating a path difference of one foot or thirty centimeters and i'll go into my plugins here and i'll take an eq which eq should which eq shall i choose occam's razor says choose the simplest so this is a standard eq that comes with pro tools but you can do this in any digital audio workstation software and i'm going to use the low pass filter the reason being that when the sound reflects from the desk even though the desk is hard there will be a certain amount of absorption of high frequencies so i'm going to put that at let me think i'm going to leave it at a thousand hertz because we do want to hear the effect and also i want the level of the reflection to be lower so let me go to the mix screen so there we have it i'll take the solo off and let me put the level 10 decibels down to account for any absorption in the desk so i'll do the same as before when the solo is on you're hearing only the original recording when it's off you're hearing the two together with that delay so let's try it i wondered lonely as a cloud that floats on high are dale and hill when all at once i saw a crowd a host of golden daffodils beside the lake beneath the trees fluttering and dancing in the breeze you see this is where it gets to a point of subtlety that i think to many listeners that would be okay but to someone who's experienced in audio then you can hear the difference the thing is if you hadn't heard the original could you hear the comb filtering without any reference to compare it with that's where it gets tricky i'll put that level back up to zero so we can hear it with the low pass filter i wandered lonely as a cloud that floats on high or dale and hill when all at once i saw crowd a host of golden daffodils yeah that's definitely degraded so that's the kind of thing that you should be listening for whenever you're recording and you see hard surfaces alarm bells comb filtering is likely to happen so you need to be listening out for it and make sure that it doesn't happen what could i do to eliminate the comb filtering from the desk they used to be in the past i haven't seen them recently so maybe they're not a thing anymore or maybe i just haven't looked hard enough you could buy acoustic tables which didn't have a solid top that had like a a lattice top so most of the sound would just go straight through and wouldn't reflect alternatively you can put something absorbent on top of the surface so just remember that sound reflects like light in a mirror the angle of incidence equals the angle of reflection or the angle of reflection equals the angle of instance as i normally like to say to get things in the right order so where it's reflecting from back into the microphone that's where you need to put your absorption and using just something as simple as a towel uh could be perfectly effective so there we have it comb filtering what fun i'm david meller course director of audio masterclass thank you for listening.

Comments on this video

You can comment on this video at YouTube

You can comment on this video at YouTube