As mentioned in another post, I travel a lot, and use my Minidisc to keep me sane. At home I keep my CD collection mainly in Wavpack lossy now, and transcode to Minidisc as I need with very good results, at least to my ears.

When I travel I generally make a few MDs for the trip, but sometimes I like to listen to other stuff, so I have been using Vorbis to keep my top 1000 tunes on my lap top, and I can transfer these to MD when I'm on the road. The thing is, the quality after transcoding to MD is OK, but not real great, compared to Wavpack with some obvious artifacts as you would expect from transcoding from vorbis 128 kbit to ATRAC3 132 kbit.

There has been an increase in interest in AAC, and some suggestion that it is superior to the others at ~128, so I thought I'd give it a try. Here are some initial findings:

I took "Enter Sandman" track 1 from Metallica's "black" CD, and "Biko" track 10 from Simple Minds' Street Fighting Years CD.

I encoded these using the following:
LAME 3.93 --preset 128 - Sandman 126 kbit, Biko 125 kbit
aacenc --streaming - Sandman 116 kbit, Biko 132 kbit
Vorbis I q4 - Sandman 118 kbit, Biko 121 kbit
wavpack lossy @265 - Sandman 264 kbit, Biko 265 kbit
musepack (v1.15r) --radio - Sandman 124 kbit, Biko 130 kbit.

Also briefly played with FAAC, but couldn't make up my mind about settings, so left it out.

I then pulled these all into foobar v0.666, replaygained by track and listened through. I also tried them all through the new ABX component in foobar, and hit 15/15 for every sample against the original wav.

At 128 kbits, you don't get transparency, at least not to my ears anyway.

Some comments about each one.

mp3: When individual instruments are playing sustained notes over other background instruments, such as in the beginning of Sandman (undistorted electric guitar) they sound slightly watery. When things start to crank up and there is more stuff going on, this gets masked and it sounds OK. Hi hats and cymbal splashes in Sandman lose definition and sound smeared. In the middle of Sandman when the quitar riff is in the thick of it, there is slight "pumping" where you can hear whooshing pumping in and out with the cranking guitars. In the beginning of Biko, there are some far off African drums rumbling quietly. In the wav, they are barely there. In the mp3, they are noticeable louder and the higher percussive instruments noticeably softer. Not sure if this is hi freq cut off, filtering, smearing/pre-echo, or a little of all three.

mp4: At first, thought that there was no wateryness in the intro sections, but when listened back to back against wavs, it's still there, but much less than mp3. Very slight pumping in the thick of Sandman, but it's barely there. The over emphasised rumbling mentioned above is not present in the beginning of Biko. Perhaps slight pre-echo/smearing, but subtle in comparison to mp3. Overall, more "transparent" in both cases to my ears than mp3.

vorbis: Not good in either case. I really expected better than this. There is a noticeable background hiss in both tracks, right from the beginning, which distracts at moderate to high volumes. In Biko intro, has similar problems to mp3, ie rumbly drums are too loud, percussive attacks are smeared and weaker in volume. In Sandman, the sustained electric guitar notes are very watery. Cymbal crashes and hihat in Sandman are very washed out. Considerably worse than mp3. Very noticeable pumping in the middle of Sandman. Disappointing.

Wavpack: Unfair comparison, just there for my own testing purposes so won't comment in detail, but it was very close to the original wav. No obvious artifacts or imbalances, but slight background hiss, much less than vorbis and not distracting.

mpc: Quite a pleasant surprise. Very similar in sound to mp4. Perhaps slightly less watery with the guitar in the beginning of Sandman than mp4, and much better than mp3/vorbis. Biko beginning is correct from a bass/balance point of view and stuff all percussive attack smearing. Some noticeable pumping right in the middle of Sandman, more than mp4, less than mp3 and vorbis.

So far, if I had to rate these based on these samples, I would pick mp4/mpc>mp3>>vorbis.

Out of mpc and mp4, mpc may be slightly less "fragile"/"watery" with solo sustained instruments, but suffers from slightly more "pumping" artifacts in Sandman.

The big test for me now though is how these perform when transcoded, which is how I intend to use them. Will let you know how I go.

Hope this is useful to others out there.

Den.

Your tests are always extremely useful, Den.

Thanks a lot.

wow, Ogg really took a beating. :'(

den > I tested some month ago different codec at 130 kbps approx (psytel --streaming, mpc --radio, lame ABR, RealOne ATRAC 132, mp3pro VBR ~130 kbps, wma9 cbr...) on a Metallica track. I listened and compared (ABX) 12 segments from One, which is quiet at the beginning, and pure metal at the end. My conclusion were approx. the same :
• 1/ musepack : some distorsions (metalic sound), but nice sound
• 2/ aac PsyTEL : quite the same, a bit more distorted, and some more pre-echo
• 3/ lame ABR : close to mp4, less details
.
I located Vorbis after these formats, especially for the high level of noise introduced by the codec, and for some less annoying reasons.

I strongly suggest you to put in the arena an QuickTime PRO encoding. For my ears, QT correct the pumping/metalisation artifacts of PsyTEL, and is closer to transparency at this bitrate that any other codec. If you have time and energy enough, test it !

Thanks guruboolez. Will look into it.

Did you ever try FAAC? I'm not real clear on the best settings for a 128 kbit sized file.

I was very surprised with Vorbis. I really thought it was better than this, but it wasn't. I am now seriously regretting encoding the 900+ Vorbis files on my lap top without doing a decent listening test first. I just tried one or two, and kinda sounded OK so I went for it. I trusted what I read on this forum. Never again...

Den.

No, I didn't tested faac. The rare approach I made were not really satisfying.

For your ~900 encodings : the most annoying artifact of vorbis (hiss, noise) isn't really noticable on common speakers. If you're not using headphone on your laptop, I don't think that you have something to regret. Have you encoded them at 128 kbps ? Conclusions may be totally different at 90 kbps : mp3/mpc are not serious competitor, and aac is not very good (for my taste) under 110 kbps. Therefore, vorbis is a really good challenger here.

I use the encodings on my laptop to transfer across to ATRAC3 on my Minidisc when I am travelling, and wish to have some new music to listen to. I always try to record ~10 minidiscs full of my favourites from Wavpack lossy files on my home PC before I leave on each trip.

Sometimes though, after about 10 days into the trip, I feel like a change, so I re-record an MD or two, from my lap top.

As you can imagine, hard disk space is limited, so I store these as 128 kbit files, and after some reading, I tried vorbis.

I know that transcoding from 128 kbit to 132 kbit ATRAC3 is really bad, but it's better than nothing when you are stuck in Japan, surrounded by Japanese pop music, and you want to listen to some Cold Chisel or INXS.

Probably the best thing for me to do is keep them on the lap top as ATRAC3 natively and avoid the transcode altogether, but then you can't do much else with them, except transfer them. I like the idea of using AAC/vorbis/mp3/musepac, because at least then I can use other players, burn them, etc.

Den.

Den, thanks for the test!

I would suggest that AAC encoder used in testing should be QTPro, because it seems to give best results. I mean you found AAC very good with AACENC, so how good it would have been with QTPro.

Hopefully NeroDigital and iTunes give AAC a nice front-end to windows. Or something similar to AAcelerator for Mac so I could easily use QTPro to encode my 400 cd's... I just like to encode once, so the encoding can take time and the encoder is allowed to cost money...

Jore, The Lazy Bastard

After many distractions, I finally got around to transcoding these further to ATRAC3 and giving them a spin in my Minidisc system.

Without going into too much detail, mp3 and vorbis could be discounted quite quickly. The original encoded files were never that great (see above comments) and the same faults were just as prevalent in the final ATRAC3 versions as you would expect. It kinda made listening for new transcoding artifacts pointless, and I soon lost interest.

I then checked out the mpc -radio and aac -streaming files onto MD as ATRAC3, and they both compared very well with the original wav as ATRAC3. The minor faults reported above were dampened some what in the transcode, and the mpc and mp4 were very close to each other in final quality as ATRAC3 files. I could still ABX them from the wav sourced ATRAC3 for each sample, but between the mpc and mp4 samples there were only very minor differences. I hate using subjective comments, but the AAC sounded a fraction "brighter", and less compressed, and both of them were just not quite as dynamic as the wav sourced ATRAC3 versions.

I then took another favourite track, Lonely Town - Stan Ridgeway, and encoded in mpc -radio and mp4 -streaming (both ~128 kbits) and then transcoded into ATRAC3 (132 kbits). In this instance. Both were missing some minor attack details in guitar sections and percussion instruments, but they were very good overall, and both very usable against my vorbis version I have used in the past, which sounds dull and less defined in comparison.

My initial thoughts are that AAC is a worthy successor to vorbis at 128 kbits as my ATRAC3 source for when I am on the road. I will test with more samples though, as musepack -radio was a surprisingly good performer as well. I still need to also check out some of the other aac encoders too, like QT 6.3 Pro when I get the chance.

I'm also ready to dabble in the AAC listening test, and will hopefully have a crack at it tonight.

I still think that Wavpack lossy is the best codec to transcode ATRAC3 from with from my home system where I can accomodate higher bit rates. B)

Den.

@guruboolez
I'm about to try out QT Pro 6.3. Do you have any recommended settings for best quality at ~128 kbits?

Thanks.

Den

The only quality settings available in quicktime are "Good, Better, Best". There's nothing else to tweak, really.

I compared a sample, encoded at BEST and at BETTER with QT 6.3 : they were identical, bit-to-bit (at least, at 128 kbps, and on one sample).

Yeah, I really expected it to do better than that as it is a good low bitrate codec, I thought MPC was good at 160k +... :'( Oh well.

Anyway, thanks for the tests. I'll be trying AAC and MPC soon in ABX, I've already tried MP3 and Ogg Vorbis at 128k(MP3 being a lot duller/blander, Ogg Vorbis being a tad brighter)

I agree.. Also I noticed that PsyTEL seemed to artificially increased the tonality of the encoded clips. As a result, it sounded far from the original audio file.