4. Understanding Aliasing - Digital Audio Fundamentals

[@akashmurthy]

I've noticed that some of you trying to replicate the audacity examples aren't getting any aliasing. Audacity is creating a 3500Hz tone instead of 7000Hz, and the debug output of nyquist says: "Warning: osc frequency reduced by 1 octaves from 7000 to 3500 hz to avoid aliasing." I've noticed the same behavior when I upgraded my version of Audacity to 2.4.2 and above. There must've been additional checks put in the code of hzosc. But if you follow the later example: return hzosc(ramp(1) * 7000) This works as expected! Aliasing does occur. So I am guessing there is a simple check to see if the number being entered is less than half the sample rate, otherwise it'll force it to be so. The reason why "ramp(1) * 7000" worked is because it's not a simple number, it's evaluated as a signal. It's a ramp signal going from 0 to 7000 in a certain duration. So I have a dirty fix for you, based on the same principle. You can try "return hzosc(pwl(0,1,1,1) * 7000)" It's not very pretty, but I'll tell you what it does. pwl is piece-wise linear function. It is essentially an envelop function. The 4 arguments state that at time 0, let the amplitude be 1, and at time 1, let the amplitude be 1. All it's describing is a straight line signal, with its value always as one. If you multiply this signal by 7000, you'll get a constant signal with 7000. Since this is not a simple number 7000, hzosc will not validate it, and will result in an aliased signal which is audible as 1000Hz. As expected. This is tested on the current latest version of 3.0.5

[@akashk961]

❤

[@arjunratheesh7743]

Wow, the sheer audacity of this guy!

[@StanleyGurvich]

Ok ive been on youtube since 2008 and this is hand down the best channel on these subjects. super simple and complicated at the same time. no bullshit. love this! This channel is going places in no time.

[@akashmurthy]

Thank you my man!

[@jitindavid2450]

Very well illustrated. You don’t find such useful info about digital audio on KZbin. Great job!!

[@akashmurthy]

Thank you for checking it out man!

[@choudharyom1]

who are you man, ur blowing things apart on YT platform, may god bless you for being so generous

[@akashmurthy]

Just a man trying to host a channel! Thanks for the words of encouragement man!

[@lcsper]

Amazing explanation. This is what I am sending people whenever I have to explain aliasing from now on.

[@carlosa.chacon985]

Totally!

[@heartwithreason]

Came here from the book: The Computer Music Tutorial by Curtis Roads in the Fundamental Concepts section where he was explaining aliasing. I am glad I read at least some of it because I found your gem of a channel! The info inside of the Roads book is somewhat antiquated (published in 95) and hard to understand but is still informative if you focus. The benefit of making this educational content in a visual and animated format, as you have done Akash, is for visual learners like myself. Keep making this intuitive content!

[@akashmurthy]

Thanks so much! I'm glad you found the channel! I'm a visual learner myself, and I tend to learn more about these concepts while I'm animating them. Thanks for the book recommendation. I need to check the book out, it's not on my reading list!

[@harshitnegi6396]

Currently watching the whole audio fundamental series and i can not thank you enough for this. An absolute gem.

[@akashmurthy]

That's awesome! Enjoy the series!

[@michaos1]

Instuctional design skills put into these videos are remarkable, amount of work is phenomenal, and it’s all on top of the great explanation and script. Bravo!

[@artsmusicstudio7379]

What a superb way to explain this concept! You have done a magnificent work Akash! Thank you!!!

[@TWEAKER01]

Great summary. To clarify one point when it comes to music production, 8:54 and 13:24 shows that when users/engineers talk of digital audio "aliasing", what they're really referring to is its byproduct: *aliasing intermod distortion* .

[@jayhu2296]

probably the most satisfying video i've watched this month, and i'm a youtube addict..

[@nathangraves4318]

-gaining foundation in material -question regarding elusive criteria -require proper conveyance specifically -has 1st and only Patreon sub

[@wergonnadoit]

KZbin is not just at all, This Chanel has a ton of amazing great information, the quality of the information and the amount of work that was put into the videos is phenomenal , I am starting a new position in an AV company, and your channel helped me a lot to understand audio i really can not do you justice

[@akashmurthy]

Thanks for the kind words! I'm really glad to know that the content helped you out.

[@princessceballos]

I came her for brief explanation but was enthralled at the science and stayed the whole time

[@akashmurthy]

Im glad you stayed! Stick around for more perhaps!

[@greatprince3190]

big thx we saw that in class and it was verry unclear, with the visual and the way you present it , it's way easyer to grasp. thx for your work

[@akashmurthy]

You're welcome, I'm glad it helped out!

[@bil63]

Great! Thank you again! I think these videos are the best explanation of digital audio aspects in KZbin!

[@akashmurthy]

Thank you for that! :)

[@GrandNecro]

i like your teaching style. lots of examples rather than just theory

[@leavy_Zola]

I am watching this as exam prep and it’s so unbelievably helpful. Thank you a lot for your work. By now I also feel like I’m getting a free Audacity introduction additionally 😂.

[@akashmurthy]

Glad to be of help! All the best for your exams

[@luiza177music]

I've always dismissed Audacity, never knew it could do that, that's awesome!

[@akashmurthy]

Yea! It's great! It now has VST3 support, which makes it a baller! But I've never thought of Audacity as a DAW, but more as an audio editor and analyzer.

[@LesVideosdOlivier]

Fantastic explanation, I feel a lot smarter now than at the beginning of the video ;) Both the explanations/pedagogy and the video aspect are awesome!

[@akashmurthy]

Thank you for the feedback, I'm glad you found it educational!

[@manikandanchembai95]

Hey yo, thanks brother. Beautiful video. Clear, concise and well presented. I'll be sure to share your videos with my peers.

[@akashmurthy]

Thanks very much bro!

[@Aleksandr-i9]

Wow! My English is so bad, but I finally understood everything about aliasing. The great explanations! Thank you so much!

[@akashmurthy]

Great stuff! Thanks for letting me know

[@boshi9]

Best explanation on KZbin!

[@cohaagenup]

Wow, this was amazing and amazingly comprehensible, even to a math-less person like me. Thanks a lot!!!

[@akashmurthy]

Thank you for the feedback! :)

[@gurratell7326]

I like your comparison with temporal aliasing in the start, people tend to not understand that aliasing is aliasing no matter the domain in a digital system, it's just perceived in different ways. Though I find it kind of interesting that it seems that it's only in the audio world that it's constantly and always killed with fire, while in movies, pictures, CG, games etc etc it's not. People even try to avoid it instead, like turning of motion blur in games, or movies nowadays more and more often have lower shutter angles (shorter shutter speeds). Of course I see why to some degree, it gets blurry, but then the only real solution is more frames or higher resolutions. I guess I just have to wait for the day when we can capture images as transparent as we can with sound today :) Anyways, you got a really good set of videos here, keep it up!

[@akashmurthy]

Thanks very much! Yea, aliasing is definitely a problem in very domain of work where discrete signals are used instead of continuous. It's unavoidable. It's just the cost doing business I suppose!

[@janihannikainen1992]

Extremely well explained and demonstrated.

[@akashmurthy]

Thanks mate!

This is really good. I'm watching all videos in this series.

[@akashmurthy]

Cheers! Thanks for checking it out. Enjoy!

[@EC-TPavithiravalavan]

Really great video and best explanation I ever seen you deserve more views

[@akashmurthy]

Thanks you!

[@altravolta]

Thanks. It was great answer.

[@XRaym]

The visualization are awesome, this is highly pedagogical, thx !!

[@akashmurthy]

Thanks mate!

[@savinaynangalia7349]

Incredible explanations 😯 Thank you so much!

[@akashmurthy]

You're welcome! Thanks for watching!

[@BharathRajeevan]

Really loved the explanation! Eyeopener, you have a new fan!

[@akashmurthy]

Thanks very much man! Glad you enjoyed it.

[@bsoach]

This is such a clear explanation - thanks Akash! Liked and subscribed

[@akashmurthy]

Thank you for checking it out! :)

[@akepamusic]

Fantastic series!!

[@liltick102]

This channel is incredible.

[@akashmurthy]

Thanks a lot mate!

[@chilledvibes5700]

Literally awesome video!!!!!!!!!! Thanks for the video.

[@akashmurthy]

Thanks for checking it out mate!

[@davidhernandezcampos1628]

Thanks Master, cheers from México

[@Orrinton]

DDMF Metaplugin and PluginDoctor are 2 of my best friends 😌☺

[@BenDinai]

Amazing Video, so clear and well presented. Thank you very much!

[@akashmurthy]

Thank you for it checking it out!

[@marcopostmartini]

Great video!

[@valevenga797]

a great video, well done, cheers

[@akashmurthy]

Thanks mate!

[@gurvinderdhiman9290]

Best Explaination 🙌♥️

[@TheZeroomg]

Was looking for video about alias but no luck, and after i'm in bed youtube algorithm do the perfect job

[@akashmurthy]

The algorithm gods are on your side today!

[@remy-]

The best on YT!

[@akashmurthy]

🤗

[@leeoiou7295]

great video

[@C622]

It is so well explained that it makes me squeal with joy! ;-)

[@akashmurthy]

That's the best expression of enjoyment I've received for any of my videos. Thanks mate!

[@ArtificialSoul]

I am trying to create soundfont preset with samples containing looped saw waves. For the ones who are familiar with soundfonts: aliasing usually occurs within such saw wave samples when they're above a certain pitch. They can sound correct when played in the same pitch as the root key, though with another pitch or even detuning with one cent aliasing occurs. So I assumed to fix it by editing the samples with a low pass filtre to remove the highest harmonics and creating "headroom" till the nyguist frequency. It doesn't help despite using a stronger cutoff (48 dB) What are the possibilities available for me to create an aliasing free soundfont?

[@danielcortesalvarez1935]

You saved my life.

[@linss7960]

Very nice video!

[@trevorbeingtrevor]

Many of your videos have brought me a greater understanding of important audio concepts. Thank you so much for your hard work! If you get a chance, I would love to know what software you use to make your excellent animations?

[@akashmurthy]

Thank you so much for your support! I use Adobe After Effects for most of the animations. Some of them were frames rendered through code in p5.js

[@kevinwydler4405]

This is so well expained!

[@akashmurthy]

Cheers!

[@ashkanaref4056]

Thanks man! Plugin Alliance is taking notes 😆

[@akashmurthy]

I don't know what that means, but sounds good! :D

[@adrianbucher5299]

Very nice explanation! I wanted to create the aliasing effect on Audacity by myself but ran into a problem. When I set sampling-rate to 8000Hz and type the nyquist command: "return hzosc(7000)", Audacity will create a 3500Hz tone instead of 7000Hz! The debug output of nyquist says: "Warning: osc frequency reduced by 1 octaves from 7000 to 3500 hz to avoid aliasing." How can I force nyqist to enforce aliasing? I'm using Audacity version 2.4.2 Many thanks!

[@akashmurthy]

Hey Adrian, thanks for checking it out. I noticed the same behavior you are experiencing when I upgraded my version of Audacity to 2.4.2 There must've been additional checks put in the code of hzosc. But if you follow the later example: return hzosc(ramp(1) * 7000) This works as expected! Aliasing does occur. So I am guessing there is a simple check to see if the number being entered is less than half the sample rate, otherwise it'll force it to be so. The reason why "ramp(1) * 7000" worked is because it's not a simple number, it's evaluated as a signal. It's a ramp signal going from 0 to 7000 in a certain duration. So I have a dirty fix for you, based on the same principle. You can try "return hzosc(pwl(0,1,1,1) * 7000)" It's not very pretty, but I'll tell you what it does. pwl is piece-wise linear function. It is essentially an envelop function. The 4 arguments state that at time 0, let the amplitude be 1, and at time 1, let the amplitude be 1. All it's describing is a straight line signal, with its value always as one. If you multiply this signal by 7000, you'll get a constant signal with 7000. Since this is not a simple number 7000, hzosc will not validate it, and will result in an aliased signal which is audible as 1000Hz. As expected. You can try posting any question you may have in the forums here. I'll surely be posting to find out if there is an alternate way to force aliasing. forum.audacityteam.org/viewforum.php?f=39

[@adrianbucher5299]

Hey Akash Awesome! Thank you for showing me a workaround to force aliasing using a constant frequency!

[@aussie_philosopher8079]

Hi akash I'm looking at getting up to speed with DSP and DAFX algorithm design in matlab (or other programs if you can reccomend) any books/information resources would he great so I can finish restart my masters at University.

[@akashmurthy]

Hey Robert, there are a couple of resources I can suggest regarding DSP. The Scientist and Engineer's Guide to Digital Signal Processing by Steven Smith - This one is free, it's available as a PDF and piecewise chapters as well. But it's a bit old and dated, and it's not fully about processing audio, it's more general signal processing, but it's very informative nonetheless. Designing Audio Effect Plugins in C++ by Will Pirkle - Probably one of the best books I know regarding audio DSP algorithms. It's very informative, very easy to understand, but the examples are in C++ not Matlab. But really, it depends on what you are planning to learn in DSP, and what your goals are.

[@akashmurthy]

Another great resource for DSP is Julius Smith from Stanford's CCMRA program. He has a few books. Amazing resources, and it's all free and available online.

[@aussie_philosopher8079]

@@akashmurthy I read this at University, it was to complicated I know about the DSP guide, that's great also DAFX from Zolzer really helped.

[@legosteveb]

Awesome video! Can you clarify… At 16:50 you summarize that they over sample then down sample. Is there a ~22kHz low pass filter step between the over sample and down sample steps?

[@akashmurthy]

Thanks! And yes, there's a mandatory low pass filter (usually an FIR filter) before downsampling a signal.

[@jakubkonarik578]

Absolutely goated!

[@bensfractals43]

how would i go about isolating this "alias"? i've seen someone else do it.

Hello again. Either, doesn't work for me when I try to make the slope signal. ¿What am I doing wrong? How do you make it work?

[@rays7805]

I get it now. This makes sense.

[@keerthana5979]

would you please clarify ,how to choose sample points for real time signal.....for example signal with sampling frequency of 360 HZ...sir ....

[@collectionofsensations]

If i record a analog synthesizer playing a square wave, using a 96kHz sample rate. That i guess would result in a sampled square wave with more higher order partials, less aliasing in the audible range than if i instead would use 48kHz sample rate. That way higher sample rates i guess is more useful in situations where aliasing would otherwise occur right? Or does the aliasing get filtered out in the ADC anyway? But the square wave in this example still needs higher resolution to get a higher fidelity sampled representation in the digital domain i suppose, that way higher sample rate would come closer to the analog domain signal? Since project sample rates affect CPU load, i am wondering when i benefit from using higher recording sample rate than 44,1kHz? If we take the square wave example, or when recording distortion effects for example, would you benefit from using a higher sample rate (96kHz) and then down sample the recording by using a lower (44,1kHz) sample rate in the project settings? To ease CPU load, and instead upsample non linear effects like compressors and saturation? 44,1kHz would create kind of edgy sampled 20kHz waves in the digital domain, but DACs will make them more rounded right through some algorithm?

[@akashmurthy]

If you convert any analog medium to digital one, you are passing it through an ADC. An ADC will filter out all frequencies above the Nyquist, whatever your sampling rate maybe. "But the square wave in this example still needs higher resolution to get a higher fidelity sampled representation in the digital domain" - No, categorically no. The sampling theorem guarantees that all frequency partials below the Nyquist frequency will be faithfully represented. "44,1kHz would create kind of edgy sampled 20kHz waves in the digital domain" - Again, no it would not, for the very same reason as above. 20kHz is below the Nyquist Frequency, so it WILL be accurately represented. The DAC does not use any algorithm to smooth it out. It uses Sinc interpolation to get back the original signal, EXACTLY as it was before being digitized. Check out video 2 on Sampling Theorem if you haven't already. "i am wondering when i benefit from using higher recording sample rate than 44,1kHz" - there is no benefit. Simple as that. As I mention in this video, the benefit is only when generating high frequency partials WITHIN the digital domain, not when recording.

[@collectionofsensations]

@@akashmurthy Million thank you Akash! 🙇‍♂

[@guntarssmits2104]

Nice educational videos with great visualization. Just wondering what tool is used to visualize your stories?

[@akashmurthy]

Thanks mate! I use Adobe After Effects for the illustrations.

[@haydenvaniderstine9645]

Superb

[@MukkuPavanKumarPHD]

Hey, thank you so much for your lott of efforts to deliver experimental and theoretical justifications about aliasing. Could you please tell me how to prepare the edits like this? Actually, I want to teach to the students like this. So I have to know how I prepare my presentation like this.

[@akashmurthy]

You're welcome! These are all individual animations that I've made and stitched together in After Effects. You can do the same if you want, or download the video, chop pieces of it and show those instead.

[@emerymesich1134]

Thank you!!!

[@largepoodle6036]

so basically, 96kHz is the optimal samplerate if you want to avoid aliasing (in conjunction with analyzing frequency spectrums to ensure there is no aliasing taking place) i would say 192khz but pc isn't gonna be able to handle that lol. i think 96khz would still be perfectly ideal if youre going to be mastering down to 48khz or 44.1, which you most likely will.

HELLO. IM TRYING TO DO IT THE SAME WAY BUT when I generate a 7000 hz didn't die me a 1000hz sine wave. Instead I get 3500. If I use 6000 it gives me 3000hz. ¿How come is this happening? What am I doing wrong?

[@akashmurthy]

Please check the pinned comment for the solution.

[@sebastianvallejoperez9754]

duuuude, how is it that you don't have more subscribers or views,

[@akashmurthy]

Haha..this is one of the better performing videos to be fair! :)

[@definty]

Nice video dude, you have a talent for teaching. Why not just use a low pass filter at the end of your plugins to stop anti-aliasing instead of oversampling?

[@akashmurthy]

Cheers! So, higher frequency output is possible in these plugins I talked about. They are a result of calculations that are unavoidable. If frequencies greater than the Nyquist frequency are generated, they are automatically aliased! So the problem here is range. If the output frequencies exceed the range, there is no point filtering them out, they are already aliased and sitting well within the range. To get rid of these higher frequencies, you'll first have to exceed the range (by oversampling) such that they don't alias! After that, you can low pass filter them.

[@definty]

@@akashmurthy Ahh I see thanks :) So you can't stop harmonics bouncing back down if your using a lower sample rate even if you use a low pass.

[@thangible]

you are awesome, thanksss

[@akashmurthy]

:)

[@andreas9483]

Great content. Do you use python for the visualisation?

[@akashmurthy]

Thanks! No, I use expression scripting within After Effects. It's loosely written in JavaScript, and it gives quite a lot of control. Sometimes, I've had to resort to a library called Processing for some visuals.

[@LakshayKukreja11]

Fuckin genius, man! Mind blown at 9:22

[@akashmurthy]

Haha! Blew my mind as well when I realised it. Cheers man!

[@deanrinehart]

How is this free?

[@deanrinehart]

Well now it’s costing me $6.50 a month… ;)

[@akashmurthy]

@@deanrinehart thanks a lot for the Patreon subscription my man! :)

[@godmeherbaba5607]

💕💕💘💘💖💖