In this recently released version of Winamp, the packaged AAC encoder has been changed!  Now the app comes with Fraunhofer encoding for HE-AAC.  Dialogs have also been changed so that the format of AAC you end up with (HE, HEv2, or LE) are set for you.  This all makes for a much nicer user experience – no need to read my previous blog for a start!  So how well does the new interface and encoder work?  As before, we look at a set of 20 tracks and clips and check out their sizes and some frequency graphs.

An improved encoder, an improved interface

In the Encoder settings for Winamp, we have a new MPEG-4 AAC Encoder!  Whoop!

Unlike the old encoder, we can pick from Variable, as well as Constant bitrate encoding, and the quality slider for each automatically changes the specific AAC encoding version that is used:

imageimageimage

This automatic AAC type selection means that even YOU, humble reader, can get the benefits of AAC-HEv2, and you don’t even know it.

There are only a few disadvantages here, but you do need to consider them, and even plan ahead to do some hair-dye-style spot checks before you start any massive re-encoding / re-synching process:

  • Not all players understand AAC-HE – as was the problem on an old build of an Android Media Player I was using, and all the sound came out dodgy

Nope, that’s it!

Method

As you will see from previous posts, we have used spectrum plots and file sizes to measure, in a “what do you reckon” type way, the performance and success of AAC and MP3 conversion.

In this post, let’s see how the encoding options here compare to options for other converters in that previous blog, and then what kind of file sizes and spectrum plots we get.  You will recall that there are 20 tracks and 20 track clips in our test block, across a range of genres.

To carry out the conversion, Winamp’s “Send” context menu is used to pass the tracks and clips to the format converter.  Here we can see that the HE-AAC encoding (but presumably not AAC-LC encoding) is done by Fraunhofer.

image

Variable Encoding

Theoretically, variable encoding is preferred (as discussed last time) as the computer can decide what bitrate is most appropriate throughout a given song and allocate as much as is necessary.  The downside is that the file size isn’t calculable beforehand, and the file sizes fluctuate.

The quality slider on the dialog uses different measures of Quality to our previous test, so it’s not easy to carry out a fair, objective comparison of this encoder against the previous or other encoders.

For example, after running all tracks through the encoder, the results show that “1” in Winamp’s new encoder reduces tracks to 2.6% of their original “pure” WAV size, “3” to around 7%, and “5” to around 15%.  Looking at our old VBR results, this puts 5 just in between the old Winamp and Nero “H” settings results, and 4 at the old “L” levels.

“1” has been completely excluded from any comparisons because we didn’t look at HE-AACv2 in any detail last time.  But I have to say, HE-AACv2 in this encoder is awesome…  For some tracks (especially where stereo wasn’t that important) it prove very difficult for friends to tell, in a blind test, which was the Q1 and which was the Q5.  Yet the file sizes are the difference between holding 10,000 songs on a 8GB phone memory card (Q1) and just 1,500 (Q5).

Variance

As in the previous tests, here is a table that shows, for each encoder’s own quality setting, the variance from that quality setting’s average compression rate across all the tracks.

image

“HON” here is used to represent the new Winamp encoder, while AAC is the old one.  As before, we do need to take care not to compare across columns, since each resulting compression average is different between encoders and encoder settings.  Some things do stick out:

  • The encoder does well when the others don’t: e.g. track 4, 7-9, and 16.
  • The encoder does badly when the others don’t: e.g. track 2, 3, 11, 15, and 18

Of course, one might argue that this could be to do with the quality setting, if it weren’t that these conclusions are across the quality settings; take track 3 for example, where it sticks out like a sore thumb at Q5 (~14% larger) and Q4 (~14.25% larger).

Spectrums

Here are some spectrum plots for some good performers (4, 7, 16) and some bad ones (2, 11, 18).

# WAV AAC-L HON-4 AAC-H HON-5
4 image
61642
image
5595
image
5267
image
10563
image
8540
7 image
49625
image
4503
image
4555
image
8605
image
7569
16 image
50490
image
4847
image
4673
image
8901
image
6095
2 image
32676
image
2986
image
3342
image
5651
image
5424
11 image
33351
image
3043
image
3465
image
5936
image
5695
18 image
48051
image
4282
image
5014
image
8686
image
8026

Remember: we can’t compare directly and say Oooooh well clearly Nero does better than Haunhofer, we can only make generalisations.  And generally, the Nero AAC Encoder from our previous tests gives good reproduction at high quality levels (e.g. Track 2, 18) and the Haunhofer does well (Track 11) too.  Generally the encoders still cut off high frequencies to save space (as we would expect).

But there are a few things that interest me: look at Track 2.  Haunhofer’s 8.5MB seems to be spent just as well as Nero’s 10.5MB – in fact there are more upper frequencies kept by Haunhofer.  It’s not the same story for 2-18 though, where a comparable amount of space is used by each, but the results look very different.

Well, we could wax lyrical about this and given my complete lack of any type of degree in physics or Marten Coltrane Supremes I’m not going to.  But as a customer, pure and simple, the Haunhofer encoder does just as well as Nero.  And, importantly, it works inside Winamp.  I can use it for ripping, and it works.  INSIDE WINAMP.  Whoot!  Finally, VBR encoding, and with a good encoder as well.  Whereas Nero was fiddly to get going in any other media player (I had to do all the conversions for the last blog on AAC using a command line interface…)

Constant Rate Encoding

I’ll be honest, we’re both bored by now.  We know CBR will give us a fixed file size, and we’re pretty much going to get a set amount of quality across the board.  What interest me at this point is whether it’s worth re-ripping out of the 160Kbps fixed rate that I used last time.

Album Old Size New Size Old Spectrum New Spectrum
The Prodigy
The Fat of the Land
65.4MB 98.9MB image image

And hey presto, the spectrum plot for track 1, “Smack My Very Naughty Girlfriend in a Tasteful Way”, looks exactly the same.  And the sound test?  Exactly the same.  Some tracks grew by around 2MB (which isn’t bad, considering the computer now had another 60kbps to play with), but a couple grew in size by over 50%.

Here’s Narayan, which grew from 10,775 to 16,743.  The left is the lossless, FLAC version of the track.  Next, the 160 CBR version, and finally our Q5 VBR version:

imageimageimage

The Q5 version seems to keep a better upper frequency representation, especially around 15,000Hz.

And here’s Climbatize, where the same is true: better upper frequencies.  And note how in both of these, a dip in the plot on the original at around 175000Hz is “smoothed over” in the 160 but not the Q5 encodings.

imageimageimage

Conclusion

Sure, Haunhofer in Winamp is a good way to rip CDs with Gracenote and album art, and a much better interface now.  And yes, on the down side, you are going to pay for this higher quality with disk space.  I can’t prove Q5 is better than 160Kbps, even though my brain says that it is: it’s not like you can’t hear people singing at 160kbps but you can at Q5.

And some people’s equipment, let alone their ears, won’t be good enough to reproduce upper frequencies with any gusto or accuracy… would they?  WELL!  These new Dr Whatsisface Headphones are spreading and multiplying like subtle mice.  Fashionable, yes, but good too.  Not only Dr’s, but also Monster headphones, and little Japanese ones that cost £50.  There may be a revolution coming.

Perhaps we are to witness the death of MP3 after all? Perhaps people will shun 128kbps MP3 downloads once they hear the glory, the majesty of Honfernenfrauer Q5 VBR AAC?  And yes, together we shall declare, “YES!  You may pay for it in disk space, but by golly, the upper frequencies really do look better on a graph than MP3 EVER COULD!”

But for me, who already has 160kbps, who already has Monster headphones and isn’t DEAD because of a lack of upper frequency plots… well… why would I bother?

Oh yes: because I’m an early-adopting neurotic that must ensure all I am and all I own is better than you and yours…

<GETS CDs FROM LOFT; FIRES UP WINAMP; MAKES CUPPA>