In a previous post on the Digital Home, we saw that AAC is the new MP3.  But is it really worth re-ripping all your CDs to AAC, and switching your purchases to AAC?  With my tiny bit of PC and hi-fi knowledge I go on a detailed journey to understand the differences between the two formats.  Come with me and see if you agree…

What’s the theory?

According to the sales pitches, AAC can give you the same quality of sound from much smaller file sizes: use up the same space on your computer and get better sound – quicker downloads, more music on your portable player…

If you’re like me and ripped those CDs you own to digital form in the 90’s, chances are you had so little disk space available you had to choose a low quality MP3 so that your CDs would fit on the 20GB hard disk that just cost you a month’s wages.

Now comes Mr AAC, with it’s better clothes and cleaner teeth.  I ask myself: is there any point re-ripping all those CDs into a new format?

What sounds better?

A typical CD is 650MB in size – that’s almost three-quarters of a gigabyte!  Ripping software, like iTunes, Winamp, or Windows Media Player, squash a CD using a “codec”.  Codecs strip out sound beyond the hearing range of most listeners, or remove sounds that cancel each other out.

We can tell codecs either to be very strict about the file of the size (bitrate compression) or we can say, hey, I don’t care about the file size – just make it sound “medium” (like video codecs). The lower the quality or the bitrate, the smaller the file and the worse the reproduction is.

Bitrate restriction is like pushing water down a pipe of a fixed diameter, restricting the amount that can be pushed down it.  Like Bill here: he only just fits in the pipe.

The theory goes that AAC squishes in a superior way than bog-standard MP3, resulting in smaller file sizes (when aim for the same quality) or better quality for the same file size (when we’re strict about the bitrate).

The tools to do it

Different companies produce different AAC codecs; as with most things, some manufacturers are better than others.  Experts with a lot more experience have already extended their extremely expressive extremities and compiled a list of different codecs, rating each one using various measures. Nero (for AAC) and LAME (for MP3) seem to come out best.

To drive a codec you need software that co-ordinates things.  It tells your computer to read music from the CD (still illegal in the UK!!) and then throw the 0s and 1s to the codec which produces an MP3 or AAC file that is then saved to your hard disk.

Here are some pieces of software, together with the codec they use for AAC or MP3.

Reader AAC MP3 HE-AAC (see later)
Fre:AC FAAC Codec LAME MP3 Encoder  
Foobar 2000 Nero AAC LAME MP3 Encoder  
Winamp Coding Technologies LC-AAC Encoder N/A Coding Technologies aacPlus (up to 128kbps)
Winamp Pro
(fast ripping)
Coding Technologies LC-AAC Encoder Winamp MP3 Converter Coding Technologies aacPlus High Bitrate (over 128kbps)
MediaMonkey FAAC Codec LAME MP3 Encoder  

Seeing the change

Audacity is a great little audio tool; one of its capabilities is to generate spectrum graphs so we can visually compare ripped tracks against the original.

Armed with 20 songs from a range of genres I created 20 second clips, converting each clip and each track into AAC and MP3.  I used three bitrates (128, 192, and 256) and three “qualities” – that’s six different spectrum graphs for every track and six for every clip.

Here are the spectrum plots for just a few of them.  You can click on an image to make it bigger.

Setting CD MP3 AAC
128kbps image image image
128kbps image image image
192kbps image image image
192kbps image image image
256kbps image image image
256kbps image image image

We can see from the plots that:

According to forum posts, the quality based approach should always be used because it follows the ebbs and flows of the music, resulting in fewer wasted bits in your files.  We should expect some much better frequency plots here than the fixed bitrate versions.

So let’s look at the Quality based encodes then:

Setting CD MP3 AAC
Low image image image
Low image image image
Medium image image image
Medium image image image
High image image image
High image image image

Yup, High Quality AAC is better than High Quality MP3.

A techie adventure into Q

I converted three tracks and took the average of the average bitrates for different VBR and Q settings to come up with VBR / Q settings for Quality as follows: Low (5; 0.4); Medium (3 / 0.5); High (0 / 0.65). It was roughly what Foobar with NeroAAC suggested anyway so I did feel a bit like a pleb afterwards.

I created spectrum plots at a range of quality settings for AAC so we could see how the sound differs at each setting when varied by 0.01 points:

Q File size / avg kbps Track 1 Track 2
.58 9170 / 220

7268 / 211

image image
.57 8962 / 215

7268 / 206

image image
.56 8749 / 209

6944 / 202

image image
.55 8541 / 204

6776 / 197

image image

I switched to another codec (Winamp).  Here are the outputs for a different track at .56 and .57:


…and for a hi-fi calibration noise:


The rule seems to be that encoding over a .57 quality setting gives a nicer representation of the original than one at just .01 lower.  Which is odd.  Anyway I can’t explain it, but I left it in here to show that I am awesome when it comes to getting obsessed with codec settings…!

Back to reality: save that disk space!

Giving the computer the power to adapt by using Quality Encoding instead of fixed bitrate encoding means you could save disk space.  With fixed bitrate, one minute is the same number of bits no matter what; with quality, the computer decides where to use all those bits up.

The spread-sheet below lists all the 20 tracks I encoded.  It shows how big each song was when quality encoded, as well as how bit it was in the bitrate version.  The bars then show how much better or worse it was in terms of file size.


Corpse Pose (Divination); Porcelain (Moby); Dancin’ (Aaron Smith); and Previous Love (Blaze / Barbara Tucker) compress better (according to the averages for each quality method) independent of whether it’s AAC or MP3, while Ghost Hardware (Burial) and I Love New York Live (Madonna) fair badly.

And which codec is more consistent when it comes to behaviour?


Of the six quality-based encoding options, MP3 compression (with this codec) comes out worse – very slightly. Notice too that the differences using AAC as opposed to MP3 are much less severe.

We’ve learnt:

  • some songs (around 20% tracks in the sample) are better encoded using quality settings rather than fixed bitrates, giving you smaller file sizes.  MP3 has the potential to be much worse or much better, though, while AAC is just a little worse or a little better.
  • AAC at bitrates over around 200kbps have the capacity to be better than MP3.  (Whether the environment you listen to your music in (tube, front room, toilet) and the equipment you have will give your ears and brain the chance to take advantage of that capacity is a different matter.)

BUT WAIT!  What this chart could be showing is the effect the Quality setting has – i.e., codecs will take advantage of our flexibility by using more space to produce better sound.  And if that is the case, AAC is able to do this more efficiently than MP3 can.  I.e., the same quality but for less cost.  Which is what Matalan apparently does, but I’ve never been in there because it smells funny and my brother says everything is CHEAP.  CHEAP!!!

Is it worth it?

It’s not a *bad* thing to want to make the stored version of music closer to the original form from the CD.  And it’s certainly convinced me to choose AAC of at least 256kbps when I buy music from a shop.  But if all that matters is that it *sounds* the same, well, none of the analysis above will tell us that.  Next then, to be sure, it’s worth comparing the experience AAC gives versus MP3 in the environment and on the equipment that will be used to listen to it.

I roped in flatmates and took 20 second samples from each track and played them back to back in a completely random order, like this one:

[Original] [MP3 High] [AAC 128] [AAC High] [MP3 128]

I’m not testing the tester’s ability to distinguish good from bad – I’m testing whether there is a noticeable experiential difference.  To that end, the listener scored the encoded tracks as either the same, better, or worse.  Then I checked patterns in the results:

Pattern One: The listener thinks AAC is better than MP3

Original >> [Worse] [Better] [Better or Same] [Worse]

Pattern Two: The listener thinks higher bitrates are better

Original >> [Worse] [Worse] [Better] [Worse]

Pattern Three: AAC at a lower bitrate is the same or better than MP3

Original >> [Worse] [Same] [Better] [Worse]

The first listener’s responses (on a 5.1 Fatal1ty Headset) gave up Pattern 3 in 48 of the 80 possible responses.  But they only thought AAC was better in 37 of the samples and that higher bitrates were better in 34.  I.e. he believes what he heard was better than MP3 if it was in AAC when the previous clip was either (a) an MP3 at a lower bitrate / quality than the previous clip or (b) AAC at the same rate.  Looked at another way, the listener failed to notice any rate changes in over 33 of the possible 60 “true test” clips, and a codec change in 36.  We’re not really getting this blinding light – this eureka moment – that we might expect.

Only 3 of the “matches” to the Pattern 3 were pure and exclusive – the others 45 occurrences could just be down to them correctly detecting a rate or codec change.

(I’ve got this data tucked away if anyone wants it!)

The second listener’s results were equally as pants.  They picked up a change in Codec or Bitrate more than 50% of the time (30 for bitrate, 32 for codec), but Pattern 3 emerged only 20 times.

Importantly, these two flatmates of mine couldn’t consistently tell the difference between AAC and MP3 – not in the way frequency plots imply.  There just is not the evidence to support a carte blanche move to AAC just because it has prettier frequency spectrum graphs.  To quote one of the listeners:

It just sounded like someone had tweaked the bass and the treble between tracks – like sometimes when there was a lot of bass, the treble wasn’t so good.  Otherwise they all just sounded the same.

AAC: The Next Generation

This entire blog entry has focused on AAC LE – the first born.  The evolved standard, formalised some 6 years later than AAC-LE, is HE-AAC.

There are three key enhancements in HE-AAC that are not part of LE:

SBR gets the high frequencies back into our recorded versions at much lower bitrates than either the original forms of LE-AAC or MP3 ever could.

PNS recognises that the sound of the sea is the same as wall-sanding. Even though they are different sound sources, PNS tells your media player to go “shhhhh” by producing white noise. This gives even more space over for other important sounds.

Finally, Parametric Stereo cleverly turns stereo music into mono, and then includes data on how to bring the stereo back. This only kicks in at ultra-low bit-rates, say 24kbps – so in our world it’s not really doing anything – but it’s great for digital radio and telephone conversations.

So, take a look at these frequency plots.  The left hand side is a standard MP3 at 128kbps – on the right is HE-AAC with SBR at just 96kbps:


And the AAC-HE file is 2MB smaller than the MP3 on the left – a whole 50% smaller.

Think back on the plots at the beginning of this blog.  We noted that, at around 192kbps, AAC-LE re-introduced frequencies at a bitrate of around 200kbps.  Given the efficiencies of PNS and SBR, will HE-AAC bring back those frequencies earlier than both AAC-LE and MP3?  Let’s look at Kelis / 4th Of July (Fireworks).  First, LE-AAC (top row) then AAC-HE (bottom row).

From left to right it’s original, 128 and then 256.



Wow! AAC-LE drops a huge block of frequencies at the top-end, while HE-AAC has near enough finished its beer and is on the way home.  Notice how AAC-LE replicates the original at 256kbps, but HE-AAC is still processing and optimising the original.

Just for a laugh, here are the MP3 equivalents for your pleasure:


It goes without saying that there is a significant disk space saving to be had, and then, necessarily, less requirement for you to have the latest and greatest network connection.  I.e., AAC-HE ought to save you money and perform better.

Let’s get into looking at bitrates, then.

Bitrate Lovely Lines File Size
96 image 4044
112 image 4707
128 image 5370
160 image 6697
192 image 8025
224 image 9352
256 image 10679

There are three groupings for this track – one including 96 and 112, one including 128, 160, and 192, and one including 224 and 256.

The last group has a fuzzier plot over 20Mhz then the middle, the middle group has frequencies over 20Mhz, and the lower group has no frequencies over 20Mhz.  At this point, you’re probably thinking, “nyeah” and would plump for 128 for this song.  Thinking ahead and outside this song (knowing that some songs do better than others) I’m plumping for 160 to give that all-important leg-room and a sick-bag for the unexpected.

By re-encoding CDs to AAC-HE 160kbps I can expect a drop in disk space usage of about 15% in moving to AAC-HE at 160kbps – two more albums on my phone for every dozen I have currently.

Let us rise up, fair readers!

In conclusion then:

  • MP3 drops upper frequencies, even at its highest settings, for every song thrown at it
  • AAC-HE beats LE-AAC at bitrates over 128kbps
  • AAC-HE does little, if anything, to sound at bitrates over 160kbps
  • AAC beats MP3 hands down in professional and beer-powered testing

In other words, I can store more and hear the same thing by switching to AAC-HE at 160kbps from MP3.  The only question now is: can I be bothered to re-rip everything?