Audio codecs & compression best books

[ycainjo]

Hi all

I would like to go deeper into audio codecs and audio file compression (not compresser fx....) knowledge.

What are there best books (or online courses) on this topic ?

[The Documentary Sound Guy]

I learnt everything I know about audio compression from Hydrogenaudio and Wikipedia back in the day.  I too would love to know if there's a more organized way to learn this stuff in book form.

Get ready for some serious math, particularly the discrete cosine transformation this is that basis for nearly all lossy data compression.

[ycainjo]

@The Documentary Sound Guy I am just discovering Hydrogenaudio, nice ressource !

Concerning the books, they have mainly been published 10 years ago or more, so nothing about new audo codecs...

Somebody talked to me about John Watkinson's Art of Digital Audio book (2000...) and Marina Bosi's Introduction to Digital Audio Coding and Standards book (2012).

I didn't read them yet...

Edited December 21, 2022 by ycainjo

[IronFilm]

17 hours ago, ycainjo said:

 

I would like to go deeper into audio codecs and audio file compression (not compresser fx....) knowledge.

What are there best books (or online courses) on this topic ?

This is really a topic that's in Computer Science, so if you wish to go deep into it expect to seriously upskill your CompSci knowledge. 

And as Computer Science is a niche within the Mathematical Sciences (most CompSci Depts grew out of originally being within the Mathematics Department), expect lots of math especially for this topic in particular. 

 

Audio codecs/compression is also heavily related to signal processing, which is very heavy on mathematics. 

 

How deep are you wanting to go? What are your goals? (come up with better audio compression??)

 

I guess I'd suggest this pathway: 

A general introductory programming course (in any language is fine, but Python is a very popular choice)

https://www.coursera.org/specializations/python-3-programming 

An introductory paper to Discrete Mathematics (do this before DS&A). 

https://www.coursera.org/specializations/discrete-mathematics

A couple of introductory papers to DS&A (data structures and algorithms. What's in DS&A is really core to progressing further, so might as well do this twice over so it gets drilled into the brain). 

https://www.coursera.org/specializations/data-structures-algorithms

https://www.coursera.org/specializations/boulder-data-structures-algorithms 

 

You've now got a solid but very basic foundation done on the CompSci side of things. But you've still not touched signal processing at all.

So take a general mathematics course in Calculus and Linear Algebra:

https://www.coursera.org/specializations/mathematics-engineers

Then do two Signal Processing courses (yes two, as again, this is really of core importance!): 

https://www.coursera.org/specializations/digital-signal-processing

https://www.coursera.org/learn/audio-signal-processing (yay, finally we got to do something specific to the audio world!)

 

Might be good to do a foundational statistics course too, either one or both of these (don't do it before the general math class though, or before taking the introductory Python programming course):

https://www.coursera.org/specializations/statistical-inference-for-data-science-applications

https://www.coursera.org/specializations/data-science-statistics-machine-learning 

 

Ok, you've now got some foundational knowledge laid down, let's get started on learning more specifically how all this can be applied to audio codecs and compression?? Well, unfortunately there doesn't seem to be many/any MOOCs specifically do with audio (well, other than the one I just linked to), but there are tonnes about the same general concepts however applied to images (as obviously image processing is a very hot topic these days!). So you could take a few of those, and a lot of the general concepts you learn would apply back to audio too.

 

https://www.coursera.org/learn/digital

https://www.coursera.org/learn/image-processing

https://www.coursera.org/specializations/image-processing 

https://www.coursera.org/learn/matlab-image-processing

https://www.coursera.org/projects/autoencoders-image-denoising

https://www.coursera.org/projects/image-compression-generation-vae

 

And beyond that, just use the background knowledge you've now gained to start learning on your own, buy some textbooks on this, and help out on open source audio projects for instance. 

 

Feel free to hit me up with any more specific questions, as while I haven't studied audio codecs and compressions specifically, I do have a mathematics degree, did a physics postgrad signal processing paper and now doing postgrad engineering analytics. 

[Dalton Patterson]

Check out the series Silicon Valley, they discuss compression in a hilarious context. 

[IronFilm]

1 hour ago, Dalton Patterson said:

Check out the series Silicon Valley, they discuss compression in a hilarious context. 

ha, I was very very tempted to make a comment in my post about middle out compression...  but I resisted the temptation to talk about D2F or T2O. Well, until now!

[ycainjo]

@IronFilm thank you for this detailled roadmap ! I have to refresh my engineer background and mathematical skills...

I will take me several months (years ?) 😉

I formerly trained in video compression and codecs, and would like now to have some knowledge in audio codecs to better understand how they work and to optimize my own audio compression parameters.

A related topic is : are there objective metrics used in audio world to assess audio encoding perceived quality (similar to PSNR, SSIM, VMAF used in video encoding) ?

Edited December 23, 2022 by ycainjo

[IronFilm]

On 12/23/2022 at 8:42 AM, ycainjo said:

@IronFilm thank you for this detailled roadmap ! I have to refresh my engineer background and mathematical skills...

I will take me several months (years ?) 😉

Depends, if you just want a bit more rough general familiarity with it, then could do that over a few weekends, but if you want to create your own new and improved codec? Might take years and years.