You are the first person I’ve run across to actually describe the difference between “reverb” and “echo.” I picked up on it eventually, but now I have an actual description should anyone ask me.
I don't understand much of audio, just clicked here to enjoy you talking... my favourite thing in this video (and what I unfortunately spent the most time paying attention to) were all the pictures on the wall. All memories actually printed and put up. It's sweet and makes me miss my friends so much.
This is probably a big part of why I often can tell that tv or movie dialogue was dubbed in a booth that had different audio properties than the original location. Great explanations!
pretty fun, i calculate impulse responses for work we usually use white noise to do it. Will try to convince my boss to buy me a gun. saw you years ago in a clip from The Technical Difficulties and man that's a picture for the 10 year challeng! one got grey, one grew a beard and the other two greatly improved their hair but didn't age a day.
I guess you could try, flip the phase of the impulse response and apply it to the sound. No idea whether this would work and/or cause some weird things to happen.
I doubt you can cleanly remove reverb this way, but I know there are standard tools to remove semi-repetitive background noise given a sample of just the noise.
@@OrigamiMarie oh yeah doing it that way would not be clean. Real spaces do stuff that a single impulse can't always account for. It would be interesting to see matt try this on some of the audio he recorded from the oil chamber to see what weird effect comes out.
Ah, removing reverb, I think you need an alchemist to do that... There's a reason that recording studios are made to be as dead to reverb as possible. It's easy to add, and a complete sod to remove!
It's been a while since my last maths lecture but I recall that convolution can act as a one way function - it's reasonably easy to do one way, but really really hard to do the other way round. That's because at every point in time, modified signal from the past and the future (!) of the recorded signal get added and superimposed over the current signal. It gets pretty, well, convoluted.
Fun tip for music producers: Some convolution reverb plugins (cough Fruity Convolver) will let you _sample_ other reverb plugins and create impulse sounds from the plugin, letting you use a .wav instead of an instance of the reverb plugin. Don't feel like paying for a big, expensive reverb plugin? Download the demo, sample it with a convolution reverb plugin, and bam! Free reverb!
You'll hopefully know this Matt - reverb is used by old-school DJs on American radio shows and one of the best exponents of the style was the wonderful Ron Sedaille at 102.9 WDRC-FM Hartford Connecticut - his airchecks are legendary and are *still* I believe on RUclips.
Have you ever tried using things other than a sharp transient in a space? I once used a contact mic’s recording of an acoustic guitar pluck as an IR, to an interesting effect. Also, there goes my afternoon - now I’ll be loading random sounds into a reverb vst.
Thank you sooooo much for the impulse response! I was going to try to grab it from Tom's video because I liked the sound so much but this is much easier :)
Thinking they should show this to undergrad engineers taking classes in signal processing. Its a very intuitive example of convolution/impulse responses. Great video Matt
Hahah, Love your face when you talk of soft furniture in the livingroom; It looks like that you for a splitsecond imagine a livingroom with hard uncomfortable furniture ^_^
You can get the impulse signal from non-impulse inputs with a bit more work. You only need to know the input signal without the reverb and the result with it. It only works if the input covers a broad enough frequency span. The trick is to do an FFT of the two cases and then effectively divide.
Matt, I love you to bits for sharing the IR sample with the rest of us! Did you use a an omni capsule or a bidirectional? I’m trying to make a preset for waves IR1 to go along with your sample.
If you have the impulse response of an general system, it is easy (for the pc) to fit an Transfer function. This function can than be applied to an different input
Now I know where the stuff I study in signal and systems comes handy in, like convolution and impulse response, as that oil tank was your system and you got your impulse response which was convolved with your voice.
As someone with both the mathsy/signal processing background, and the sound engineering background, this is fascinating! It completely didn't occur to me to use the DSP signal processing impulse response to modify a recorded sound to apply reverb. That's a really cool idea!
Thank you!!!! Ever since I heard that, I wanted the impulse (I was thinking it was called profile). And now that I know what my DAW wants to create custom reverbs, I think I'm going to capture impulses in a few of my favorite places. :)
This is a really good video. You explained a complicated process in lay terms and made it understandable to a wider audience. I really enjoyed this and learning how faking audio is possible when required. I'd love to see more of these in future.
Super excited about these.. As a filmmaker (sort of), I really struggle sound-wise.. Can't wait for more explainy-tutorialy content about sound and stuff!
Another cool thing about a convolution reverb is that you can back up speakers to remember what they sound like, so even after the speaker is no longer in your possession, you can make you music sound like it's being played through the speaker that you once had.
You generally want a sine sweep from 1Hz to 48000Hz, so you don't "miss out" on any of the frequency responses, because there's different dropoff for different frequencies when they have to "around" (tiny) surface angles
An impulse has all the frequencies in it so the impulse works You don't need to go to 48KHz. Almost no recording has anything about 20KHz. FM radio only does about 15KHz. Most videos you will find cut off below that.
@@kensmith5694 you grab 48khz mostly to have that data available for processing, not so much for the final audio. Kind of the same way you want 14 bit raw for photography, not just the 8 bits that the final JPG or PNG will be in. While the final audio file will typically get cut off after 12 or 15KHz, having everything up to 48 gives you much more data for things like pitch shifting.
@@jurrich It there is no energy at or near 48KHz, there is not need to know the gain and phase at that frequency for the task at hand. Other cases are a different question bit in this one, you could safely only have information up to about 15KHz. Beyond that, you can just make the gain number zero and not worry about the phase. BTW: I know of a case where someone got data, applied a huge phase vs frequency shift and then processed the data then applied the reverse phase shift. This allowed him to use a a fast but low resolution processor to make a very high resolution (spike like) data output.
Pardon my ignorance. Can I, or someone else, use this to create one of those “*song* but playing next door/down the hall” type videos? Is this what would be used to make “*song* but played in massive oil tank” video?
A single impulse response wouldn't get you all the way; it's great at capturing the reverberation addd to a sound source "down the hall" or "played in massive oil tank", but doesn't reproduce how you'd need to attenuate different frequencies of a sound from "next door" or "oil tank that is actually filled with oil". However, convolutions are really powerful and niftily versatile: If you don't use an impulse response as the convolution kernel (the kernel is whatever is convolved with the source audio to compute the result of the convolution), but instead a different mathematical function, the same algorithm can act as a frequency-specific filter (e.g. low-pass or high-pass). So even though convolutions aren't omnipotent in terms of digital signal processing (e.g. they can't do everything you'd need to model a guitar amplifier), I think you could implement the effect you're thinking of with a small chain of convolution effects.
@@nex Any linear frequency attenuation (any 'eq') can be reproduced using a kernel. Also a chain of convolution effects can always be baked into a single kernel just by convolving one kernel with all others. This isn't enough for a guitar amp though since they usually have plenty of non-linear effects.
This was a great explanation and tutorial! What exactly was the preset you loaded - did it just define where those sliders should be, or something more?
Just moved the sliders to tweak it to sound a little better, and adjusted the wet/dry mix so you could hear a bit of the original audio too not just the sound of the room.
Aww hell ya.. Matt just sitting talking to camera about things he's interested in would make for an amazing series!!! Even things he's not an expert at... Like.. Swimming in a bee suit... And its... Effects on the environment.
Very cool, cheers Matt! I remember SoundBlaster soundcards back in the 90s(?) that allowed you to select whether you wanted to sound like you were in a concert hall or a cave etc. Now I'm wondering whether that team just wanted to fire guns all over the world :) Thank you for the implied warning about the gunshot, I have tinnitus from not being aware enough about noise induced hearing damage to protect my hearing when I was young, so that was nice :)
Awesome!!!! Would you share the full recording you used for the convolution reverb for us mortals that couldn’t go and play the eeriest flute and kazoo in the world?
I used to be an audio technician in the 1960s and 1970s. We had a "bedspring" reverb in the studio, which would make rather a plain singer with a plain acoustic guitar sound a whole lot fuller - until someone bumped into the rack, and everything got all boingy and weird. (Sometimes we wanted it boingy and weird - this was the Sixties.) Now how does that compare with convolution?
In a room like that, they would sound terrible. With such a long reverb everything just gets muddy and hard to distinguish. Notice how even in this video the last few words he speaks with the impulse are already starting to get overcrowded by the reverberation, and that's with him tweaking the wet/dry mix to make it sound better.
@@suzannep Both probably. Long reverb makes music sound super weird because notes will clash with the notes that come before and after. Also remember organs are almost always in places with lots of reverb, so this would be an increase in quantity, but would not be a qualitative change.
If you have a nice organ (like you might find in cathedrals and some churches) it'll probably have been designed for the space it's in, so it wouldn't sound as good elsewhere necessarily.
My question was going to be "Can we get the IR you recorded?" but then you answered it... :p I'm currently doing a PhD in sound localisation that has me playing with a motion tracking system, a 24-channel loudspeaker array (in a ring) and occasionally some IR generation tools (like Odeon) in a hemi-anechoic chamber. It's great fun :)
Cool! I remember someone posting a comment on ... I think it was maybe an episode of Citation Needed saying they just wanted videos of you explaining things. I am enjoying your uploads, Matt. Can you make more videos, please? I am not too fussed on what they are about or whether they make sense.
Good explanation, thanks! Would be interesting to hear what's the difference between something recorded in the "the real" location and the same thing "faked" with convolution reverb.
you know when they are measuring reverb and stuff you know it gets quieter slower the quieter it gets so at some point you have to say this reverb too X amount of seconds or whatever, if you use more sensitive equipment it just gets harder to measure. i think they should just have reverb as a half life which is to measure stuff like decay or whatever since there's no real point where it just goes away completely
The maths is a lot more interesting than the actual thing imo. A convolutional operator is used a ton in signal processing; basically it is a thing where you do the integral of two wavefuntctions multiplied together with the arguments slightly changed. This, along with its inverse, is how you can fit so much in an ethernet cable.
Please for the love of gosh make more of these kinds of videos! I'm a musician (I know how to play the piano and ukulele, plus maybe a tiny bit of guitar and 10 string kantele) and I'm super interested in this! You just basically explained to me why the videos I've shot at home, of just me playing, sound terrible. If I had the energy to learn video and audio editing, I'd try and make myself sound better that way. Because I'm lazy, and have access to an entire university campus just across the street from me, I can just reserve a lecture hall to play in if I need to.
This convolution reverb has a problem: The aplitude. While the loud bang is used for convolution, it has another frequency and phase responses than a normal talking voice. And for say, while 100 dB sound event has a echo time of 100 seconds, a talking voice such as 60 db, has much less reverb time and different impulse responses. An approach could be to load different samples in different dynamic ranges, recording them in the range of the loud bang, to a whistle of a mouse. Then mix the output according to its RMS level. I tried some of these impulse loaders for the convolution with my didgeridoo but it sounded always a bit to unnatural and too "high-frequent". But well explained. I also like your other videos of inchindown, this place really interests me.
@@SibaNL For precise results you need a good room acoustic simulation, but i dont know any software for now. Computation of the spreading of the soundwaves, the sound field, could by done by ray tracing. But Iam looking also for good products and solutions to "imitate" to be in the Inchindown oil tanks or in any other places where I could design a custom reverb
@@aeroDidge Yeah, that interested me as well. You can recreate a room in something like blender but I wouldn't know about software to simulate sound in that room.
Technically you don't need to use a reverb sound in the effect. I don't have any means of trying this myself, but, you can import a random audio file and use that as the "reverb".
The mathematical theory behind a convolution reverb is essentially to just play the convolution sound file (the convolution kernel) at the level of each incoming sample of audio, once for every sample, with each previous copy of the kernel sustaining over all of those that follow. this brute force method essentially mimics the effect of having a speaker and a microphone occupy the same point in space where the real life microphone was.since a continuous waveform is basically just a series of infinitely short transients at a known level strung together (each sample is a bang at the right volume) and you're recreating how the space responded to a bang at that level for every bang, you re-create the sound of that space.now in real life it isn't done this way because it would require absurd amounts of RAM and computational power to keep piling more and more copies of the kernel sustaining on top of each other, so some fancy math I don't understand is used.
also for more professional applications, you also need to have an impulse response of the room the sound you want to convolve was recorded in, which is then """deconvolved""" from the audio before the new kernel is added.I.E. you have to remove the "bedroom" part of "matt in the bedroom" before adding the "inchdown" kernel, otherwise what you get is actually "matt in the bedroom inside inchdown" which isn't quite the samealternatively you can put lots of overpriced foam panels around yourself to dampen the impulse of the room as close to a dry transient as you can get it.
Apparently, stairwells can be great places to record when you want plenty of reverb - at least that’s where my teacher recommended recording brass instruments (though preferably after normal school hours, when the building would be pretty much empty).
I think this is my first "sit down and talk to the camera" type video. Hopefully it made some sense!
Matt Gray it was good, very easy to understand and informative
More pls
I can't speak for anyone else, but as an audio engineer myself it made perfect sense 😎
@@robburgess4556, as a non-audio plumber who dabbles, I'd say it made sense! Definitely more, please, Mr. Gray! I do enjoy your speaking bits!
it was very good, informative and friendly
I just put a snare through this reverb and it was breathtaking. It could spark a whole new genre...Inchinwave?
Vaporinchin :P Limitless possibilities!
You are the first person I’ve run across to actually describe the difference between “reverb” and “echo.” I picked up on it eventually, but now I have an actual description should anyone ask me.
Please make more of these. It’s like Tom’s videos but simpler, and using Matt’s amazing way of describing things
Enjoyed this, thanks Matt, was slightly distracted by enjoying your photos!
I was disappointed by the lack of That Face from That Video though.
*any song* but its played in an oil tank
Nerix Start with Frere Jacques on recorder. No offence, Matt...
Next viral hit. Matt, get on it!
We need All Star played in an oil tank.
Bee Movie - but played in an oil tank
Is it too late for Redbone?
I don't understand much of audio, just clicked here to enjoy you talking... my favourite thing in this video (and what I unfortunately spent the most time paying attention to) were all the pictures on the wall. All memories actually printed and put up. It's sweet and makes me miss my friends so much.
Aww. That's such a cosy picture wall, how nice!
David_From_TheDVIChannel I love how it has his friends on it
This is probably a big part of why I often can tell that tv or movie dialogue was dubbed in a booth that had different audio properties than the original location. Great explanations!
It's funny that after I had seen Tom's video, my first thought actually was to use the gunshot in a convolution reverb.
Love the pictures on the wall.. all the sweet memories of the past.. 😀 especially the one in the Matt mobile..
Yeah I was thinking that!
pretty fun, i calculate impulse responses for work we usually use white noise to do it. Will try to convince my boss to buy me a gun.
saw you years ago in a clip from The Technical Difficulties and man that's a picture for the 10 year challeng! one got grey, one grew a beard and the other two greatly improved their hair but didn't age a day.
I’m enjoying it more than i expect to be honest. Please do more stuff like this.
I have updated the download link in the description for the impulse responses!
can you do the reverse? removing the echo using the impulse response?
I guess you could try, flip the phase of the impulse response and apply it to the sound. No idea whether this would work and/or cause some weird things to happen.
I doubt you can cleanly remove reverb this way, but I know there are standard tools to remove semi-repetitive background noise given a sample of just the noise.
@@OrigamiMarie oh yeah doing it that way would not be clean. Real spaces do stuff that a single impulse can't always account for. It would be interesting to see matt try this on some of the audio he recorded from the oil chamber to see what weird effect comes out.
Ah, removing reverb, I think you need an alchemist to do that... There's a reason that recording studios are made to be as dead to reverb as possible. It's easy to add, and a complete sod to remove!
It's been a while since my last maths lecture but I recall that convolution can act as a one way function - it's reasonably easy to do one way, but really really hard to do the other way round. That's because at every point in time, modified signal from the past and the future (!) of the recorded signal get added and superimposed over the current signal. It gets pretty, well, convoluted.
I love how you think and say things- probably because it's how I think. Please think and say more things. thanks!
Fun tip for music producers: Some convolution reverb plugins (cough Fruity Convolver) will let you _sample_ other reverb plugins and create impulse sounds from the plugin, letting you use a .wav instead of an instance of the reverb plugin. Don't feel like paying for a big, expensive reverb plugin? Download the demo, sample it with a convolution reverb plugin, and bam! Free reverb!
Matt Grey, the reverb-iest man on RUclips.
People get Delay and Reverb both mixed up and it is easy to understand why. I love reverb in most cases, it is very interesting to play around with.
As someone just taking a course on digital signal processing, I find very neat that I get why the hell an impulse response is all you need
You'll hopefully know this Matt - reverb is used by old-school DJs on American radio shows and one of the best exponents of the style was the wonderful Ron Sedaille at 102.9 WDRC-FM Hartford Connecticut - his airchecks are legendary and are *still* I believe on RUclips.
Have you ever tried using things other than a sharp transient in a space? I once used a contact mic’s recording of an acoustic guitar pluck as an IR, to an interesting effect. Also, there goes my afternoon - now I’ll be loading random sounds into a reverb vst.
Thank you sooooo much for the impulse response! I was going to try to grab it from Tom's video because I liked the sound so much but this is much easier :)
Thinking they should show this to undergrad engineers taking classes in signal processing. Its a very intuitive example of convolution/impulse responses.
Great video Matt
Hahah, Love your face when you talk of soft furniture in the livingroom; It looks like that you for a splitsecond imagine a livingroom with hard uncomfortable furniture ^_^
Modernist minimalism.
YES I'm so glad you included the responses. Was the first thing I hoped for when seeing you were with Tom in the tank. Thanks a lot!
Thanks for the download! As a musician, on the initial video I thought how cool it would be to record something there, thanks to you, now i can.
That was a cool thing to watch. Thanks for making it, Matt!
I really liked this video. It shows how much you like what you do (working with sound, of course).
You can get the impulse signal from non-impulse inputs with a bit more work.
You only need to know the input signal without the reverb and the result with it.
It only works if the input covers a broad enough frequency span.
The trick is to do an FFT of the two cases and then effectively divide.
This is really cool Matt! I really want to see more videos like this !
I was talking to the audio producer group I'm in about the IRs, I'm so happy you got these! :D
Matt, I love you to bits for sharing the IR sample with the rest of us! Did you use a an omni capsule or a bidirectional?
I’m trying to make a preset for waves IR1 to go along with your sample.
It's recorded with a Røde NT-4, which is a stereo pair of cardoid mics.
Awesome! Excited to see more of you.
My first thought when I watched that video was "I NEED IR'S for this NOW". Thank you!!
Dang, I gotta get that impulse response - when I create music, I basically drown EVERYTHING in reverb :P
love the way he's enjoying reverb's thing! Keep up great work man 2020 still usefull :D
If you have the impulse response of an general system, it is easy (for the pc) to fit an Transfer function. This function can than be applied to an different input
Now I know where the stuff I study in signal and systems comes handy in, like convolution and impulse response, as that oil tank was your system and you got your impulse response which was convolved with your voice.
As someone with both the mathsy/signal processing background, and the sound engineering background, this is fascinating! It completely didn't occur to me to use the DSP signal processing impulse response to modify a recorded sound to apply reverb. That's a really cool idea!
Thank you!!!! Ever since I heard that, I wanted the impulse (I was thinking it was called profile). And now that I know what my DAW wants to create custom reverbs, I think I'm going to capture impulses in a few of my favorite places. :)
So what would it be like if you combined your convoluted reverb with the slow-mo of your lips flapping?
It wouldn't really make sense, because reverb isn't slow mo, it's more like a abnormally long shadow.
What's making sense got to do with anything?
Combine this reverb with that sound from that video of that face.
I think you wanted to say That Image from That Video?
This is a really good video. You explained a complicated process in lay terms and made it understandable to a wider audience.
I really enjoyed this and learning how faking audio is possible when required. I'd love to see more of these in future.
I want more of these please more
This was great! Good explanations and it was easy to understand! I'd love to see more videos like this.
Amazing and simple to follow explanation of the reverb stuff! Thanks for sharing the sample AND the method of using it with us. :)
I'm studying signal convolution right now in Uni that is super helpful for intuition
Super excited about these.. As a filmmaker (sort of), I really struggle sound-wise.. Can't wait for more explainy-tutorialy content about sound and stuff!
This was awesome. I'd love more videos like this for sure
Thanks Matt! I really enjoyed this video
Another cool thing about a convolution reverb is that you can back up speakers to remember what they sound like, so even after the speaker is no longer in your possession, you can make you music sound like it's being played through the speaker that you once had.
So some more videos like this matt there great
Man, we need so much more Matt sound engineering videos!
That was so cool! I never knew there was a way to add reverb from a previous recording.
Love this kind of shortish audio tutorial.
You generally want a sine sweep from 1Hz to 48000Hz, so you don't "miss out" on any of the frequency responses, because there's different dropoff for different frequencies when they have to "around" (tiny) surface angles
An impulse has all the frequencies in it so the impulse works
You don't need to go to 48KHz. Almost no recording has anything about 20KHz. FM radio only does about 15KHz. Most videos you will find cut off below that.
@@kensmith5694 you grab 48khz mostly to have that data available for processing, not so much for the final audio. Kind of the same way you want 14 bit raw for photography, not just the 8 bits that the final JPG or PNG will be in. While the final audio file will typically get cut off after 12 or 15KHz, having everything up to 48 gives you much more data for things like pitch shifting.
@@jurrich
It there is no energy at or near 48KHz, there is not need to know the gain and phase at that frequency for the task at hand. Other cases are a different question bit in this one, you could safely only have information up to about 15KHz. Beyond that, you can just make the gain number zero and not worry about the phase.
BTW: I know of a case where someone got data, applied a huge phase vs frequency shift and then processed the data then applied the reverse phase shift. This allowed him to use a a fast but low resolution processor to make a very high resolution (spike like) data output.
Broadcast Engineers are AWESOME!
What DAW do you use? I don't recall seeing options in FL Studio or Audacity for convolution reverb ^-^"
I really like this type of video from you. Are you going to be making more of these types of videos?
Very interesting, a good insight into how you do your thing, and well presented too. Keep it up!
Awesome video! I subscribed due to your musical experiments, and now I look forwards to more!
Pardon my ignorance. Can I, or someone else, use this to create one of those “*song* but playing next door/down the hall” type videos? Is this what would be used to make “*song* but played in massive oil tank” video?
yeah, those... remixes (?) are pretty much just a lot of reverb and a bit of equalising, plus the sound of rain or whatever added
A single impulse response wouldn't get you all the way; it's great at capturing the reverberation addd to a sound source "down the hall" or "played in massive oil tank", but doesn't reproduce how you'd need to attenuate different frequencies of a sound from "next door" or "oil tank that is actually filled with oil". However, convolutions are really powerful and niftily versatile: If you don't use an impulse response as the convolution kernel (the kernel is whatever is convolved with the source audio to compute the result of the convolution), but instead a different mathematical function, the same algorithm can act as a frequency-specific filter (e.g. low-pass or high-pass).
So even though convolutions aren't omnipotent in terms of digital signal processing (e.g. they can't do everything you'd need to model a guitar amplifier), I think you could implement the effect you're thinking of with a small chain of convolution effects.
@@nex Any linear frequency attenuation (any 'eq') can be reproduced using a kernel. Also a chain of convolution effects can always be baked into a single kernel just by convolving one kernel with all others.
This isn't enough for a guitar amp though since they usually have plenty of non-linear effects.
This was a great explanation and tutorial! What exactly was the preset you loaded - did it just define where those sliders should be, or something more?
Just moved the sliders to tweak it to sound a little better, and adjusted the wet/dry mix so you could hear a bit of the original audio too not just the sound of the room.
@@MattGrayYES Thank you for the extra explanation :)
i love your channel so much
Can it simulate the reverb growing and shrinking as if it's being recorded from very far away?
Pretty neat! I love this sort of video. I don't even do anything related to audio but I love to learn things about it anyways
Played for 30 seconds in audacity. Didn't see convolution reverb but did see reverb settings for bathroom and church hall
Doesn't come with audacity out of the box, you need to install a plugin for it: wiki.audacityteam.org/wiki/GVerb
Thank you for uploading the impulse response.
Thanks for impulses!
Aww hell ya.. Matt just sitting talking to camera about things he's interested in would make for an amazing series!!! Even things he's not an expert at... Like.. Swimming in a bee suit... And its... Effects on the environment.
Very cool, cheers Matt! I remember SoundBlaster soundcards back in the 90s(?) that allowed you to select whether you wanted to sound like you were in a concert hall or a cave etc. Now I'm wondering whether that team just wanted to fire guns all over the world :)
Thank you for the implied warning about the gunshot, I have tinnitus from not being aware enough about noise induced hearing damage to protect my hearing when I was young, so that was nice :)
love your work
At risk of sounding super weird, Matt has a really nice voice, gave me slight ASMR to listen to this :o
he should start an asmr channel... GraySMR
Can’t wait to play with this when I finish my shift; speed metal but the drums and bass are in a monolithic oil can? The possibilities!!
This is cool! I've got a bit of background in Signals and Systems (just a course at uni) and this made perfect sense!
I'd say it worked quite well!
Awesome!!!! Would you share the full recording you used for the convolution reverb for us mortals that couldn’t go and play the eeriest flute and kazoo in the world?
Really interesting and informative, thank you!
I used to be an audio technician in the 1960s and 1970s. We had a "bedspring" reverb in the studio, which would make rather a plain singer with a plain acoustic guitar sound a whole lot fuller - until someone bumped into the rack, and everything got all boingy and weird. (Sometimes we wanted it boingy and weird - this was the Sixties.) Now how does that compare with convolution?
Awesome video! really enjoyed it
random one you no how organs ( musical type) sound epic i wander how epic they would sound in that place ?
I'm guessing either amazingly awesome, or unbearably painful to hear. \(°~°)/
In a room like that, they would sound terrible. With such a long reverb everything just gets muddy and hard to distinguish. Notice how even in this video the last few words he speaks with the impulse are already starting to get overcrowded by the reverberation, and that's with him tweaking the wet/dry mix to make it sound better.
@@suzannep Both probably. Long reverb makes music sound super weird because notes will clash with the notes that come before and after. Also remember organs are almost always in places with lots of reverb, so this would be an increase in quantity, but would not be a qualitative change.
If you have a nice organ (like you might find in cathedrals and some churches) it'll probably have been designed for the space it's in, so it wouldn't sound as good elsewhere necessarily.
@@AdamLloyd98 OK so we need a dreadful carnival organ and make it sound great ;)
My question was going to be "Can we get the IR you recorded?" but then you answered it... :p
I'm currently doing a PhD in sound localisation that has me playing with a motion tracking system, a 24-channel loudspeaker array (in a ring) and occasionally some IR generation tools (like Odeon) in a hemi-anechoic chamber. It's great fun :)
Cool! I remember someone posting a comment on ... I think it was maybe an episode of Citation Needed saying they just wanted videos of you explaining things. I am enjoying your uploads, Matt. Can you make more videos, please? I am not too fussed on what they are about or whether they make sense.
you can get better results with different impulse types. sine wave sweeps work quite well
Good explanation, thanks! Would be interesting to hear what's the difference between something recorded in the "the real" location and the same thing "faked" with convolution reverb.
you know when they are measuring reverb and stuff you know it gets quieter slower the quieter it gets so at some point you have to say this reverb too X amount of seconds or whatever, if you use more sensitive equipment it just gets harder to measure.
i think they should just have reverb as a half life which is to measure stuff like decay or whatever since there's no real point where it just goes away completely
So when I joked that you should have brought a grand piano rather than a recorder, you could say, “ah that was an outtake!”.
Now we just need someone to redo an episode of Citation Needed with one of these.
The maths is a lot more interesting than the actual thing imo. A convolutional operator is used a ton in signal processing; basically it is a thing where you do the integral of two wavefuntctions multiplied together with the arguments slightly changed. This, along with its inverse, is how you can fit so much in an ethernet cable.
Nice! More Matt explains a science thing using unscience words!
Please for the love of gosh make more of these kinds of videos! I'm a musician (I know how to play the piano and ukulele, plus maybe a tiny bit of guitar and 10 string kantele) and I'm super interested in this! You just basically explained to me why the videos I've shot at home, of just me playing, sound terrible. If I had the energy to learn video and audio editing, I'd try and make myself sound better that way. Because I'm lazy, and have access to an entire university campus just across the street from me, I can just reserve a lecture hall to play in if I need to.
I have the biggest, platonic crush on you.
This convolution reverb has a problem: The aplitude. While the loud bang is used for convolution, it has another frequency and phase responses than a normal talking voice. And for say, while 100 dB sound event has a echo time of 100 seconds, a talking voice such as 60 db, has much less reverb time and different impulse responses. An approach could be to load different samples in different dynamic ranges, recording them in the range of the loud bang, to a whistle of a mouse. Then mix the output according to its RMS level. I tried some of these impulse loaders for the convolution with my didgeridoo but it sounded always a bit to unnatural and too "high-frequent". But well explained. I also like your other videos of inchindown, this place really interests me.
You can use frequency swoops I guess
@@SibaNL For precise results you need a good room acoustic simulation, but i dont know any software for now. Computation of the spreading of the soundwaves, the sound field, could by done by ray tracing. But Iam looking also for good products and solutions to "imitate" to be in the Inchindown oil tanks or in any other places where I could design a custom reverb
@@aeroDidge Yeah, that interested me as well. You can recreate a room in something like blender but I wouldn't know about software to simulate sound in that room.
Technically you don't need to use a reverb sound in the effect.
I don't have any means of trying this myself, but, you can import a random audio file and use that as the "reverb".
The mathematical theory behind a convolution reverb is essentially to just play the convolution sound file (the convolution kernel) at the level of each incoming sample of audio, once for every sample, with each previous copy of the kernel sustaining over all of those that follow. this brute force method essentially mimics the effect of having a speaker and a microphone occupy the same point in space where the real life microphone was.since a continuous waveform is basically just a series of infinitely short transients at a known level strung together (each sample is a bang at the right volume) and you're recreating how the space responded to a bang at that level for every bang, you re-create the sound of that space.now in real life it isn't done this way because it would require absurd amounts of RAM and computational power to keep piling more and more copies of the kernel sustaining on top of each other, so some fancy math I don't understand is used.
also for more professional applications, you also need to have an impulse response of the room the sound you want to convolve was recorded in, which is then """deconvolved""" from the audio before the new kernel is added.I.E. you have to remove the "bedroom" part of "matt in the bedroom" before adding the "inchdown" kernel, otherwise what you get is actually "matt in the bedroom inside inchdown" which isn't quite the samealternatively you can put lots of overpriced foam panels around yourself to dampen the impulse of the room as close to a dry transient as you can get it.
why does youtube delete paragraph spaces now. this is terrible
I like this kind of video, I like explainy videos. This works well for you Matt
This was a really interesting video. I would love it if you had other topics like this you could do stuff on
Waiting on that bandcamp link for the unamplified version....
Apparently, stairwells can be great places to record when you want plenty of reverb - at least that’s where my teacher recommended recording brass instruments (though preferably after normal school hours, when the building would be pretty much empty).
Don't do it at the 1/2 way point. That doesn't sound very good.
Ken Smith
The way I remember the stairwells at that particular school, the halfway point wasn’t the most tempting place to record anything.
Found out you can do this with Matlab been messing around with it.