Comparison AAC vs MP3 @ 128k and 96k

[mr.muscade]

Hello,

I searched comparison MP3 vs AAC with a method of audio spectrum but i didn't. So, i did my own test and comparison @ 128k and 96k. From a sample of reggae @ 44 Khz, i encoded to several MP3 and AAC. Encoding tested : LC-AAC, HE-AAC v1, HE-AAC v2 and MP3. I used audacity to draw spectrum. Spectrum data exported to excel and regrouped to make 1 graph. From my results, here my conclusion :

LC-AAC 96k = MP3 96k
HE-AAC v1 96k is better than MP3 128k
HE-AAC v2 64k = MP3 128k

Here, spectrum of my conclusion :
http://3.bp.blogspot.com/_lshatTNzmOg/SaPTqJfZdxI/AAAAAAAAC5A/TKsLW1A7ikY/s400/compare_excel.jpg (http://3.bp.blogspot.com/_lshatTNzmOg/SaPTqJfZdxI/AAAAAAAAC5A/TKsLW1A7ikY/s1600-h/compare_excel.jpg)

If you want to see more of my comparison then you can download my excel data. The graph is dynamic, you can select samples to compare. You can download all my excel data and samples encoded, in the middle of my french blog, here :
http://memo-mediacoder.blogspot.com/2009/02/comparison-mp3-vs-lc-aac.html

;-)

[Dark Shikari]

Er, you know that the frequency cutoff is an encoder decision that has nothing to do with the format, right?

And that the frequency cutoff may be set to a lower value because doing so would increase perceived quality at a given bitrate?

And that there is no remotely useful mathematical or algorithmic method of determining the relative quality of two audio streams?

[mr.muscade]

oh... i didn't known.... so my comparison is bad !
lol

anyway, thank you for this precisions :-)

[mr.muscade]

Er, you know that the frequency cutoff is an encoder decision that has nothing to do with the format, right?

And that the frequency cutoff may be set to a lower value because doing so would increase perceived quality at a given bitrate?

And that there is no remotely useful mathematical or algorithmic method of determining the relative quality of two audio streams?

I trust you, but could you show some sources ? Because we can find a lot of comparison on the web with this method of spectrum audio. I'm confused.
:confused:

thank you

[nurbs]

I have no sources for you, but I thougth that was common knowledge when it comes to audio encoding. An important part of lossy audio encoding is deciding what you can actually hear and what not, so the encoder will drop or encode less accuratly stuff that is masked by other tones. Also if it was as simple as looking at a spectrum why would people bother doing double blind listening test which take a lot of participients and time.

[skromnibog]

Your method has big mistakes.
Human perception of audio quality is very complex. It can't be mapped to some simple measure nor to some simple graphs.
Your method doesn't even take into consideration real difference between original and compressed signal. So, one could just create encoder which would replace original signal with noise that has the same average frequency response as original signal, and your method would claim that it is perfect.

For objective measurement of differences you can check PEAQ (ITU-R BS.1387). There are also other measurements. There were also good graphical representations of differences between compressed and original audio here: http://www.hydrogenaudio.org/forums/index.php?showtopic=1095 but those graphs are not available anymore.

Simple checking of spectrum (especially of average spectrum) is not good method of checking audio quality.

[mr.muscade]

In consideration of your all explained, should i delete this topic ?

[kypec]

Why would you want to delete this topic? Let it be for future inexperienced visitors who might learn a lot from it. That's the purpose of all technical forums IIRC.