SonicStage 3.3 ATRAC3/ATRAC3plus unscientific codec tests

[dex Otaku]

Doing some listening tests tonight to see how the atrac3/plus codecs in SS 3.3 fare.

Note that my comparison with older versions of the codecs is from memory, and thus highly fallible.

This is note an AB/X test or anything even mildly scientific in nature. It is simply me switching between tracks that I know the properties of, and comparing them by ear.

Take these results with a large grain of salt.

I used only one test track: "Railwayed" from the 1990 album Strange Free World by Kitchens of Distinction [band from Wales]. I chose this track because I've been using it for test purposes since the early 1990s when I got the album [original CD]; it is sonically very dense [high activity across the spectrum all at once] and has a heavy [but not excessively bright] high-end that very easily pushes equipment like A/D and D/A converters, makes plainly audible distortion in players that have bad synchronisation, clocking, low-pass filtering, or are experiencing high read error rates or jitter. It's absolute hell for lossy codecs to encode; encoders of any bitrate are pushed right to their limits in trying to do a decent job with the full audio bandwidth, let alone the densely-packed high-end of this track.

I have used this same track to run bitrate comparisons in SS before, but not since v2.3.

Encoding was done on my Athlon 2500+ (barton) with Windows XP SP2, and 512MB of RAM.

Tracks compared using my RH10 and Sennheiser HD330s, EQ disabled:

* PCM reference track, copied from CD by SS

* 48kbps "normal", copied from CD by SS

* 48kbps "high quality", copied from CD by SS

* 64kbps "normal", copied from CD by SS

* 64kbps "high quality", copied from CD by SS

* 105kbps [no quality setting available] transcoded from PCM

* 132kbps [no quality setting available] transcoded from PCM

* 256kbps "normal", copied from CD by SS

* 256kbps "high quality", copied from CD by SS

* 352kbps [no quality setting available] transcoded from PCM

* 48kbps-base "high quality" AAL transferred in normal mode

* 352kbps transcoded from 48kbps-base "high quality" AAL

Results, keeping in mind that this is totally unscientific and personal opinion based on my own perception:

* The PCM track sounds as expected, and reveals no obvious errors in extraction from the CD.

* 48kbps is substantially better than the last time I did this. There is obviously a great deal missing from the sound.

* 64kbps is improved as well. The difference between 48 and 64 is noticeable in there being less missing from the high end.

* I would still choose not to use either HiLP mode for portable listening [as they both give me the feeling that I'm listening to low-bandwidth internet radio], but either would likely be great for in-car listening where ambient noise levels are relatively high and playback fidelity is often compromised in many ways.

* difference between 48kbps and 64kbps HiLP in "normal" and "high quality" modes is not obvious in any way.

* 105kbps sounds substantially better than it did before, but that's not really saying much to me. I would still choose not to use it, as it still feels like sand being poured into my ears.

* I find both 48kbps' and 64kbps' artifacting less annoying than 105kbps, even though 105kbps has a much more present high end.

* 132kbps also sounds substantially better than it did before, but is still very "granular"; I might consider using it for some content, and feel it would be perfect for in-car listening.

* 256kbps sounds transparent compared to PCM; high-quality mode sounds no different.

* 352kbps sounds transparent compared to PCM; next to 256kbps there is no difference to my ears in quick testing.

* 48kbps-base AAL is the same as other 48kbps, as expected

* oddly, 352kbps-trancoded from 48kbps-base AAL sounds slightly clearer, but is still obviously coming from the 48kbps source, making use of AAL completely useless in my opinion - unless you use SS for general listening from your PC.

* There was no noticeable difference in encoding times between "normal" and "high quality" modes, despite "normal" also being labelled, "(faster)"

Totally Unscientific Conclusions:

* 352kbps isn't sufficiently different from 256kbps to warrant using SS to transfer tracks for general listening.

* I'll still be more likely to use HiSP or PCM than 352kbps, unless 352kbps is made accessible through Simple Burner.

* The listening difference between normal and high-quality modes isn't substantial, but the encoding times aren't different enough to warrant using normal over high unless your machine is very slow.

* LP2 sounds sufficiently better compared to my last tests with it that I'll likely use it to make discs for in-car listening in the future.

* AAL is completely useless. If they made it so SS would transcode from the lossless copy to a chosen bitrate, it might have some use.

* I'll continue using Simple Burner to copy my CDs for portable listening

[The Low Volta]

while I do fully understand that these results are completely unscientific... they are also exactly what I found after some rough tests myself so just to chip in my thoughts:

"indeed, dex!"

[dex Otaku]

An addition:

AAL is basically useless unless you use SS for regular listening - better than AAL support would be enabling support for all installed directshow audio codecs so that any track in a format the user has installed a codec for [including FLAC, WavPack, APE, shorten, and other formats widely used by recordists for lossless archival] can both be played and transcoded to a3+ formats within SS.

Arguments against that would seem to be that it would widen the door to copying pirated music to your portable, however - anyone with greater-than-newbie computer experience can figure out how to convert whatever they want to WAV and import it into SS for transcoding.

Also, the fact that MP3 support is built-in, as well as unprotected WMA support, completely negates any possible logic in this thinking. The fact that what is purported to be the most widely-used encoding format for piracy is already supported, MP3, takes care of this quite neatly.

MP3 support in SS already uses installed directshow codecs. I use ffdshow's mp3lib support with 24-bit decoding and noiseshaped dither with no problems whatsoever and have done so for about a year - it even lets me apply decent EQ to SS's mp3 playback if I so wish.

The only real issue is that metadata support would present issues for the SS programmers. This is also a near-null argument; most other audio formats use only a few variants of tagging schemes [the most prominent being id3v2.x and APEv2] as can be exemplified by the excellent tag-retrieval and editing support in other programs like Foobar2000.

Sony deliberately exclude support for other file formats, despite their legitimate use in recordist circles, lossless-packing formats in particular. Sites like archive.org carry hundreds of 100% legal downloads of recordings - many of which were originally made with MD or HiMD recorders - in formats like shorten, WavPack, and FLAC. It's pretty ridiculous that they should exclude support for them when it could be as simple as allowing the program to recognise their file extensions and enabling support for directshow filters that are either purchased, free, or open-source - the enabling of which will infringe on no IP laws whatsoever.

[ynos]

what about the 160 and 192 kbps?

[ozpeter]

One of the issues with any lossy format is the consequence of multiple encode/decode cycles (eg in broadcast transmission chains, or if you use the format for re-editing rather than just for delivery.

It would be interesting to see what happens if you re-encode the various ATRAC options say 10 times, which will quickly emphasise the inadequacies of each and put their merit/demerits beyond doubt. Obviously I could do that myself when time allows....

[dex Otaku]

what about the 160 and 192 kbps?

I could only test bitrates supported by HiMD.

[Christopher]

Good summation; would've been interesting to hear if there were variances on other music but that doesn't take away from the review or anything. Does anyone else not detect a different between normal and high quality modes? Perhaps if we organize a consensus..

[Sparky191]

I feel I can hear a difference between HiSP and 352kps from a WAV.

Copy 2 copies of the same track one HiSP and the other 352. Let some one else select which track to play say 8 times. with each track being played four times. But in any order. See if you can tell them apart 8 out of 8 times. Haven't tried it myself.

[Christopher]

It is also a shame we cannot record with 352kbps ATRAC3plus via the units. It better not be like that next gen.

[The Low Volta]

Copy 2 copies of the same track one HiSP and the other 352. Let some one else select which track to play say 8 times. with each track being played four times. But in any order. See if you can tell them apart 8 out of 8 times. Haven't tried it myself.

sparky... just get winABX and you can do all this with statistical anlysis and even more... and better, you do not need any friends for it

[greenmachine]

I didn't even try 352 kbps yet since i can't even abx Hi-SP from the original and if i could it would be very, very hard. For me, 352 seems just like a waste of disc space, although some 'golden ears' might hear the difference.

[dex Otaku]

re: 352kbps -

Golden ears or not, I'd use it for recording were it available.

For portable listening I find HiSP sufficient. The only time I tend to use PCM is when copying things such as Bach's Toccatos & Fugues which are all pipe organ recordings, which I find most forms of lossy compression tend to make sound raspy or even gargly.

[Sparky191]

sparky... just get winABX and you can do all this with statistical anlysis and even more... and better, you do not need any friends for it

I'll check it out. But placebo effect or not, I like what I like. HiSP is fine for most things. Its only at night or in a quiet office when you'll really hear the difference with 352. Also I tend to listen the same couple of albums for couple of weeks, before I feel the need to change them so I don't need a shedload of tracks with me. Which is why despite having a large 60GB+ music collection my Shuffle, and Zen Micro aren't maxed out.

Edited November 16, 2005 by Sparky191