These have been the best Audio series I have ever seen. Not just on RUclips, but anywhere else, including university. Thank you so much for making them and sharing with the world.
Man.. started watching your audio fundamentals and now I'm hooked on your entire channel. you're god send. pls keep doing what you're doing for people like me :)
Hey Akash 👑, your videos are simply the best on youtube, please continue and teach us more about this whole analog\digial audio signal processing world, it's so interesting and amazing how you can simplify these topics which take my uni profs a few lectures to half arse teach. I was wondering if you can share how you make your videos, are there certain programs that you use (like manim, after effects,etc)?
Thanks so much for the kind words! I am planning to continue the digital audio fundamental series soon enough. Just need to finish off this particular series. I just use After Effects to make the videos. I learnt most of the things needed to make it via a course on udemy. I think it was called Motion graphics and data visualization.
@@akashmurthy Can't wait for more videos! Oh wow, that's impressive! Only using After Effects, I can't imagine how long it takes you to create & edit each video, kudos to you, brave master!
Akash sir, would you like to make a tutorial video about ambisonic and spherical wave and explain if any relationship between in practical application.. Thank you.
No! Not by a long shot. I've just put this on the back burner for a while, since this series has been wildly unpopular! The last few videos were going to be all the ways you could misuse floating point format in the context of audio. I'll get to it at some point of time.
For those who curious why difference between "lowestNormalNumber" and "secondLowestNormalNumber" is so small (at 05:21). In previous lessons we learned concept of "window" and "offset" where "window" is interval defined by "exponent" e.g. from 2^2 to 2^3 = from 4 to 8. And "offset" defined by mantissa which interpreted as "divide window interval into a lot of equal parts". For 32 bit number mantissa occupy 23 bit which means window can be divided into 2^23 parts. For simplicity assume that there are only 2^3 = 8 parts. So our interval 4..8 can be divided into parts of size "0.5": 4, 4.5 ... 7, 7.5, 8. Well, now back to main question: when I watched video for the first time I thought that variable "precision" which is calculated as: "secondLowestNormalNumber" - "lowestNormalNumber" should be equal to the size of "window's part" i.e. to the value of "lowestNormalNumber" itself, because "window" divided into "equal" parts, so it's big enough to be represented by normalized number, right? No, the key here, is that window interval starts from "lowestNormalNumber"(which is 2^-126) and ends at 2^-125. Therefore interval from 0 to "lowestNormalNumber" is not part of the "window". In terms of previous example, where window is 4..8 you can treat "0" as 2 or 3 whose are before 4, so they are out of the scope. As of the difference between "lowestNormalNumber" and "secondLowestNormalNumber" it's really small 1.4012...E-45 and it is same for all parts i.e. "3rd lowestNormalNumber" - "2nd lowestNormalNumber" also = 1.4012...E-45. The "lowestNormalNumber" itself is a "big number", as well as 4 is a big number in comparison to a part of the 4..8 window. I hope this TLDR will help someone.
thank u for the great video honestly i'm just learning about how floating points are represented in the memory and i'm just trying to visualize where NAN is ?? hhhhh i know it sounds crazy but from the values we have -NAN should be somewhere before -infinity and +NAN should be somewhere after +infinity right ? sorry for commenting outside the topic 🙃
You're welcome, glad you like the content! NaNs are not numbers, so they can't be represented visually on a number line. They are just a flag to indicate that the floating point format contains garbage value in memory that can't be mapped to any number based on the rules of the floating point format.
Sir these are just pure gem💎 100x better than what we get to learn in colleges 🙌 I have a request to you sir, Pls... Pls upload videos related to image signal processing too. 🙏🙏🙏
@@akashmurthy Generative adversarial Network it is a generative neural network which help to generate a real copy from training data usually it is used in images but images also provided in sequential numerical format what i wanted to do is why don't we try this with sound with respective to nyquist condition so neural network will create new sound using our sound dataset there's a special methods for audio but i wanted to make it with image network, will it possible to make it and will it reconstruct form that gan output. i got this idea when i saw your sampling video which help me to understand more deeply. Thanks!
Yes this is what colleague lesson should be.As an audio major student, thank you so much. BTW I'm bit confusing about digital EQ processing, IIR,FIR,Filter phase.Will you do some lesson about these in the future?
Thanks very much! Yes, those are quite tricky topics to talk about from first principles, and to explain it in intuitive ways without using a lot of math. I definitely have them in mind!
Awesome series! It must have taken ages to plan, record, visualize and cut everything... I've watched countless videos and read quite a lot about C++ and audio fundamentals in the past few weeks since I want to get started with JUCE. Today I binge watched pretty much every video your channel, and I have to say this is one of the best resources I have found so far. Each video is thoughtfully structured, and there's a whole bunch of useful and interesting stuff condensed into relatively short videos without becoming overloaded or boring... How you visualize everything not only has a modern & professional look to it, it also tremendously helps to grasp complex concepts in a clear and easy-to-follow way. Thank you for sharing all this! I'm looking forward to more content, especially if related to FIR / IIR coefficients, Convolution / IR cab sims, amp simulation and stuff like that :-)
Thank you very much for that breakdown! I'm really glad you found these videos digestible. I'll hopefully be doing more DSP topics later down the line.
Corrections:
7:36 The sign bit should be zero if the number is positive.
These have been the best Audio series I have ever seen. Not just on RUclips, but anywhere else, including university. Thank you so much for making them and sharing with the world.
That's super kind words! Thanks very much mate.
This is the greatest video on number representation I saw anywhere, thank you thank you thank you!
Hey that's awesome! Thanks for the comment, there's a whole series of this if you're interested in the topic.
Gold on RUclips ...I'm super happy I found Your channel. Thank You :)
I'm glad you found the channel! Thanks!
Man.. started watching your audio fundamentals and now I'm hooked on your entire channel. you're god send. pls keep doing what you're doing for people like me :)
Thanks mate! I will be bring out new videos. But I'm curious, do you find this series useful? If you do, where do you apply this knowledge?
currently binging all of your youtube courses... THANK YOU!!!!!!!!!!!!!
You're welcome! Happy binging!
Awesome Series!
Hey Akash 👑, your videos are simply the best on youtube, please continue and teach us more about this whole analog\digial audio signal processing world, it's so interesting and amazing how you can simplify these topics which take my uni profs a few lectures to half arse teach.
I was wondering if you can share how you make your videos, are there certain programs that you use (like manim, after effects,etc)?
Thanks so much for the kind words! I am planning to continue the digital audio fundamental series soon enough. Just need to finish off this particular series.
I just use After Effects to make the videos. I learnt most of the things needed to make it via a course on udemy. I think it was called Motion graphics and data visualization.
@@akashmurthy Can't wait for more videos!
Oh wow, that's impressive!
Only using After Effects, I can't imagine how long it takes you to create & edit each video, kudos to you, brave master!
@@Tydox it takes a bit of time for sure! 😅 But I'm getting my animation process more streamlined and things are not as slow as you'd imagine!
Akash sir, would you like to make a tutorial video about ambisonic and spherical wave and explain if any relationship between in practical application.. Thank you.
Is this series done? Would love to see videos circling back to audio-specific applications.
No! Not by a long shot. I've just put this on the back burner for a while, since this series has been wildly unpopular! The last few videos were going to be all the ways you could misuse floating point format in the context of audio. I'll get to it at some point of time.
Would love to watch them too! @@akashmurthy
7:36 Shouldn't the "sign" bit equal to 0?
Good catch! I'll pin it as a correction.
For those who curious why difference between "lowestNormalNumber" and "secondLowestNormalNumber" is so small (at 05:21). In previous lessons we learned concept of "window" and "offset" where "window" is interval defined by "exponent" e.g. from 2^2 to 2^3 = from 4 to 8. And "offset" defined by mantissa which interpreted as "divide window interval into a lot of equal parts". For 32 bit number mantissa occupy 23 bit which means window can be divided into 2^23 parts. For simplicity assume that there are only 2^3 = 8 parts. So our interval 4..8 can be divided into parts of size "0.5": 4, 4.5 ... 7, 7.5, 8. Well, now back to main question: when I watched video for the first time I thought that variable "precision" which is calculated as: "secondLowestNormalNumber" - "lowestNormalNumber" should be equal to the size of "window's part" i.e. to the value of "lowestNormalNumber" itself, because "window" divided into "equal" parts, so it's big enough to be represented by normalized number, right? No, the key here, is that window interval starts from "lowestNormalNumber"(which is 2^-126) and ends at 2^-125. Therefore interval from 0 to "lowestNormalNumber" is not part of the "window". In terms of previous example, where window is 4..8 you can treat "0" as 2 or 3 whose are before 4, so they are out of the scope. As of the difference between "lowestNormalNumber" and "secondLowestNormalNumber" it's really small 1.4012...E-45 and it is same for all parts i.e. "3rd lowestNormalNumber" - "2nd lowestNormalNumber" also = 1.4012...E-45. The "lowestNormalNumber" itself is a "big number", as well as 4 is a big number in comparison to a part of the 4..8 window. I hope this TLDR will help someone.
good stuff!
thank u for the great video honestly i'm just learning about how floating points are represented in the memory and i'm just trying to visualize where NAN is ?? hhhhh
i know it sounds crazy but from the values we have -NAN should be somewhere before -infinity and +NAN should be somewhere after +infinity right ?
sorry for commenting outside the topic 🙃
You're welcome, glad you like the content!
NaNs are not numbers, so they can't be represented visually on a number line. They are just a flag to indicate that the floating point format contains garbage value in memory that can't be mapped to any number based on the rules of the floating point format.
Sir these are just pure gem💎 100x better than what we get to learn in colleges 🙌 I have a request to you sir, Pls... Pls upload videos related to image signal processing too. 🙏🙏🙏
Thanks for the kind words mate! But I only talk about what I'm familiar with, and I ain't familiar with image signal processing.
6:06 You fiend! How dare you!? Good day to you sir!
I, sir, have done worse.
impresionante los videos, muy buena explicacion, me encanataron, felicitaciones
Is it possible to use GAN on sampled audio
What's GAN?
@@akashmurthy Generative adversarial Network
it is a generative neural network which help to generate a real copy from training data usually it is used in images but images also provided in sequential numerical format
what i wanted to do is why don't we try this with sound with respective to nyquist condition so neural network will create new sound using our sound dataset
there's a special methods for audio but i wanted to make it with image network,
will it possible to make it and will it reconstruct form that gan output.
i got this idea when i saw your sampling video which help me to understand more deeply.
Thanks!
I'm not familiar with machine learning and its usage in audio. So, I can't add to this.
Yes this is what colleague lesson should be.As an audio major student, thank you so much.
BTW I'm bit confusing about digital EQ processing, IIR,FIR,Filter phase.Will you do some lesson about these in the future?
Thanks very much! Yes, those are quite tricky topics to talk about from first principles, and to explain it in intuitive ways without using a lot of math. I definitely have them in mind!
Awesome series! It must have taken ages to plan, record, visualize and cut everything...
I've watched countless videos and read quite a lot about C++ and audio fundamentals in the past few weeks since I want to get started with JUCE. Today I binge watched pretty much every video your channel, and I have to say this is one of the best resources I have found so far.
Each video is thoughtfully structured, and there's a whole bunch of useful and interesting stuff condensed into relatively short videos without becoming overloaded or boring... How you visualize everything not only has a modern & professional look to it, it also tremendously helps to grasp complex concepts in a clear and easy-to-follow way.
Thank you for sharing all this! I'm looking forward to more content, especially if related to FIR / IIR coefficients, Convolution / IR cab sims, amp simulation and stuff like that :-)
Thank you very much for that breakdown! I'm really glad you found these videos digestible.
I'll hopefully be doing more DSP topics later down the line.