OSNews

I don't think that the guy has any "agenda" regarding any of these things. He is a software engineer student and he uses Linux AFAIK.

Ya, because we all know that Linux-using software engineers never have an agenda.

Edited 2005-12-15 01:09

Eugenia, you should mention this article to members of hydrogenaudio.org and see their response. It is very likely that they will recommend you to pull down the article due to the unscientific nature of its testing methods.

fully agreed. i see no ABX results, no other than these graphs - how are they created? it seems no real listening tests are done, only digital analysis. that is not audio codecs should be tested.

and they fail to mention ogg vorbis, musepack and other high(er) quallity codecs.

They are neglecting the fact that in MP3 and AAC SNR has _NOTHING_ to do with quality, as both are _PERCEPTUAL_ codecs.

What a terrible loss of somebodies time that PDF is.

If the mp3 didn't produce high/low end very well re-encoding it wouldn't help. If you're saying that it just "sounds" better, obviously the data's in the mp3(because it was in the AAC), mabye it's the software or codec you're using to listen to it.

I can only hope the guy means that he re-ripped them from CDs. Re-encoding mp3s->aac them would make them sound worse in almost every case and this guy just made a complete ass out of himself.

Ya, I've noticed this in ripped mp3 -> ogg vorbis converts, they sound awfull compared to a ogg vorbis cdrip.

You re-encoded the mp3s? Then you have the losses of mp3 and aac. And someone cut your left ear with a chainsaw, but you didn't noticed. ;p

Em... Transcoding from one lossy format to another will only serve to decrease quality. No, transcoding the file to wav first won't help you either as your MP3 encoder has permenantly "damaged" the original stream. The only way to do this properly is to archive the CD track in lossless formats (FLAC, Apple lossless, Wavpack) and transcode from there.

Also, there is no way for you to distinguish the quality difference at such high bit rates (disregarding the format and encoder being used) without professional equipments, a pair of "golden ears" and double-blinded listening tests. You should be safe staying with MP3 at 192bit.

how should an aac file sourced from a mp3 file you described as "bad" sound better than the source ? Where should the needed information comae from ?

I actually like Ogg, it does sound cleaner and "sweeter". unfortunately, it has almost no hardware support or commercial download support.

You know, vorbis was updated since v1.0 and there are people actively tuning the encoder...

From what I've seen, a good AAC codec (the study I saw suggested Nero AAC) can beat Vorbis, but Vorbis is definitely better than MP3 (This I know from testing myself) and better than low to middle end AAC encoders. I think Apple's AAC codec varied between better and worse than Vorbis. I really wish I had the link... The guy had tested different musical styles, different sound styles through Ogg Vorbis 1.1, Lame 3.96(or latest, whatever that was), and several AAC codecs.

All I can say for this study is, if LAME MP3 beats FAAC AAC, FAAC must be an absolutely terrible encoder. If AAC can beat Vorbis, and Vorbis is better than MP3, AAC should be better than MP3.

As for AACplus, I'm pretty sure from my own experience that AACplus is MUCH better than MP3 and Vorbis at low bitrates. I have no idea how it does at higher bitrates because I haven't seen any AACplus files encoded at higher bitrates. They probably don't exist because the whole POINT of AACplus is its performance at low bitrates.

EDIT:
Wait a second, Will Entriken! I know him! We both go to Villanova... He lived in my dorm two years ago...

You know him? Then go and kick him in the head for doing such a _stupid_ test. Sheesh. There've been approximately seventy six zillion comparisons of lossy audio codecs done in the last five years, and ALL of them have been better than this. As someone's already pointed out, you simply can't compare a single metric which is rather irrelevant to perceptual encoding and declare one codec the winner. The only way you can really usefully compare codecs is to use a variety of music and do controlled subjective tests. People who do this kind of test with the correct methodology usually come up with very consistent results that show the best encoders for every format (mp3, wma, vorbis, aac...) very close together, with usually aac at the top of the pile by a tiny amount. And in the end it's fairly bloody academic anyway, because very few people have equipment good enough to tell the difference between codecs when listening portably (hint: ipod earbuds don't cut it), and anyone who cares about quality uses a lossless format for desktop listening. rant ends.

I completely agree. While I cannot condone such a violent action against any person, this test is very unscientific. You should tell him to pull down the article, apologize to the FAAC developers and learn more about audio encoding first.

you might to take this one, or the current hydrogenaudio.org test:

http://www.soundexpert.info/coders128.jsp
http://www.hydrogenaudio.org/forums/index.php?showtopic=39448

From what I've seen, a good AAC codec (the study I saw suggested Nero AAC) can beat Vorbis, but Vorbis is definitely better than MP3

What do you mean by "what I've seen"? Any links to support your comment? Because in every blind (or double blind) test I've seen vorbis beat aac and mp3 hands down, by a fairly large margin. (this is just one example: http://www.infoanarchy.org/story/2002/9/8/23472/23921) AAC better than mp3? It depends what test you look at... I assume they are similar in capabilities ... AAC is probabely a better format - I guess gapless playback is not a problem like with mp3... Oh, and ogg supports some nifty features like bitrate peeling, gapless playback, multiple channels, and it is faster than mp3. How fast is aac compared to mp3 btw?

ogg is better than mp3 in low bitrates. musepack is the best of the pack in high bitrates.

I don't know, but I was quoting a different test (the precursor to the new HydrogenAudio double-blind listening test that several people have recommended).

only a "few" players already support it
http://wiki.xiph.org/index.php/VorbisHardware

Volkswagen and Volvo recently added support for it in their new car audios.

It isn't about leet-ness and hacking. It is a little about freedom and a whole lot about sound quality and file size.

By the way, the excellent Rio Karma player supports ogg and so do some of the samsung players, which are also great.

No good players support it? Maybe support is limited on the hardware side, but I doubt Foobar2000, Winamp, AmaroK, Rhythmbox, XMMS, BMP, XMPlay, and Deliplayer are all bad players. No, Windows Media Player and iTunes don't support it (out of the box), but it seems like everything else does, on PC at least.

Conclusion. ITunes and Windows Media Player are not very sophisticated tools....

Conclusion. ITunes and Windows Media Player are not very sophisticated tools....

Nonsense. iTunes and Windows Media Player’s failure to support Ogg Vorbis has nothing to do with their level of sophistication. There are two reasons why they do not support Ogg, and neither of the reasons have to do with the technical merits of the programs. First, Ogg Vorbis just isn’t very popular. It’s a very well designed codec, but it doesn’t have a large enough following that Apple and Microsoft feel compelled to offer native support in their respective media players. It’s very rare when a large for-profit corporation spends time and resources supporting a product that offers them zero chance to increase their profits. Second, Apple and Microsoft have competing audio codecs in AAC and WMA that they want to establish as the de facto standard. Neither of them can ignore MP3 because it’s been entrenched for so long, so they include support for MP3 along with their own codecs, but there is no way they are going to support Ogg Vorbis unless people start asking for it in large numbers. If millions of people started demanding Ogg Vorbis support, it would pop up in iTunes and WMP about a week later, along with a firmware update for the iPod to support Ogg files.

Edited 2005-12-15 17:19

Ogg Vorbis is widespread and can be used in WMP, if you want to mess around with getting it to work. It doesn't support it out of the box, but nor does WinAMP actually.

It's already possible to use Ogg Vorbis in WMP and many other media players on windows. Just install the proper codec and spent some time on tweaking apps like WinAmp and WMP. It's not intuitive what needs to be changed here and there.

Ogg Vorbis is in very high demand though mp3 is still the ruler.

Winamp does support vorbis through a plug-in that's installed by default during a full install of the player. Winamp has had this support since 2.8x or so.

Well it's there, but it didn't work for me on Win2K nor on Win2K3. It doesn't work out of the box. Some configuring of WinAMP is needed, incl. installing a codec utilizing DirectSound. The plug-in did not work on it's own, which was quite weird.

Em... Now who said there are no good software player to support vobis? Any Quicktime/Directshow player would support it (with the QT and Directshow filters from Xiph)! This includes WMP and iTunes!!

Well, aren't you a little closed-minded lout? So you think the only 'good' digital music player is the iPod? And the only good software player is iTunes? You must be a Windows user, because no Mac user would talk that way. Either that, or you're about 14 years old, and have no experience whatsoever outside your little bubble. You'll be surprised to learn that there are DOZENS of players out there, all with sound quality equal to or greater than the iPod, and software players (like WinAmp, XMMS and amaroK) which play MP3s, OGGs, RealAudio, and even (gasp) WindowsMedia Audio.

Grow up, or get a life, and stop bashing others who don't think the way you do. There's way more than one way to listen to music.

can anyone recommend a good portable msic player? so far only iRiver and Cowon iAudio devoces play FLAC and Ogg/Vorbis at a decent quality.

I have an iAudio G3 and a friend of mine has an X5. I would recommend either one of these, especially if you listen to a lot of spoken word stuff (podcasts, audiobooks, etc). Bookmarking is good

PS - Not recommended however if you want to patronize the online DRM-laced music stores (which I don't recommend you do). There's currently a firmware upgrade that plays these files, but it's currently in beta. In any event, I think it is still a better buy than iPod in this regard, unless you plan to only stick with ITMS. (WMA burned to CD and then ripped back to mp3 [which you would have to do to play on an iPod] sounds like ass. Not that WMA files sound stellar either way, but it sounds even worse after a lossless conversion to another format.)

Edited 2005-12-15 04:12

WMA burned to CD and then ripped back to mp3 [which you would have to do to play on an iPod] sounds like ass. Not that WMA files sound stellar either way, but it sounds even worse after a lossless conversion to another format.

WMA to MP3 is not an example of a lossless conversion.
Besides that, the quality of the resulting file depends on the quality of the original and the quality of the encoder and settings used for the resulting file. In the example given, you're going from one lossy format (unless you're using WMA Lossless) to another -- cutting out information with each conversion.

Yeah, I didn't mean lossless .. it was a typo

I use my Palm Zire 31, and apart from a little bit of his from its slightly noisy preamp, it has tremendous frequency response. So if you have a PDA or smart phone, consider using that as well.

There is always Rio Karma and Samsung produces some Vorbis-capable players. However, I do not recommend playing lossless files (ie. FLAC and Apple Lossless) on portable music players. They will only serve to shorten the battery life of the player with no perceived quality increase since you are very likely to use a non-professional output device (ie. anything that is not $900+) in a noisy environment (actually, it's more like any environment that is not completely silent), which will make it impossible for you to distinguish any minuscule quality difference.

I'm a happy Rio Karma owner, however I would not recommend it futurewise. The reason is simple really, Rio Audio has ceased supporting the product and has left the Mp3portable industry. They're shutting down operations as we speak and you won't see much new things happening there.

If that is not an issue, Rio Karma might still be a choice....

Well, Rio Karma is a kickass player that handles FLAC and Ogg. Though since Rio is out of business now, they might be tricky to get hold of.

Rio Karma, my personal favorite. Ogg/Vorbis, FLAC, mp3 etc. etc. It has, however, been discontinued (apple did a great job forcing everyone out of the market) but if you can find one second hand somewhere or from rio audio's site it's worth a look.

I'm confused why the measurement is SNR. Typically SNR is audio signal vs. background noise and in the 80 - 120 dB ballpark. Deviation of audio signal from desired signal is normally measured as a THD percentage, I thought. How do we interpret a ~30 dB SNR of audio signal from desired signal deviation?

Well, THD is a tad more complex than just "the deviation from desired signal" (nor that it is wrong, it's just more complex in reality - there are many factors involved in calculating THD aka Total Harmonic Distortion).

Before I answer your question I'll have to reread the PDF again. Usually SNR is used in the context you (and I) clearly expected.

Actually, SNR is completely useless when it comes to comparing between lossy encoders. Lossy encoders make use of the human ear perception and try to add as much "noise" (actually, it's more like modifying the stream) as possible without having the "noise" detected by the human ear. Thus, the ONLY way to evaluate lossy encoders is to have a group of volunteers do ABX (double-blinded listening tests) on a variety of samples.

I agree on that. Especially the part about double-blinded listening tests. They are a necessity no matter what kind of audio equipment you are testing, incl. audio codecs.

"LAME is, on the other hand, the best implementation of MP3 currently"

Correct me if I'm wrong but I thought LAME is a copy of Fraunhofer's mp3 from a time when also the source was available.
If true then Fraunhofers current version is probably better
http://www.all4mp3.com/tools/sw_fhg_cl.html

No. You're completely wrong.

At one point LAME was a set of patches against the MP3 reference encoder. That's not the same as Fraunhofer's encoder - it was simply a basic encoder that generated valid MP3s, although it did so very slowly, and the resulting quality was far from optimal. Terrible would be a better bet. The original LAME patches improved the reference encoder significantly. After a while, the entire encoder was rewritten from scratch, and doesn't even operate on the same principles anymore. It uses a completely difference psychoacoustic model, for example.

It's much like Apache in that regard. Apache was originally a set of patches against the NCSA web server. Since then, it has been completely rewritten.

Fraunhoffer's MP3 encoder is nowhere near as good as LAME.

The Greatest advantage to the new Fraunhofer (Fhg) mp3 encoder (from all4mp3.com) is that it is about 5 times faster than the LAME encoder. The Fhg code has been optimized by Intel and AMD.

on my system (3.2 Ghz P4 Extreme) the new fhg encoder encodes at 79x, lAME encodes at 17x.

my benchmark file was a 27:39 wav file, Fhg encoded it in 21 seconds, LAME encoded it in 98 seconds.

well, if you want to stick to old technology and not use the best mp3 encoder you're free to use fhg.

No. It's removed all the dist10 reference source code. It's still distributed as source code.

see http://wiki.hydrogenaudio.org/index.php?title=LAME

Oh, I know it's heavily codec- and song-dependent. My point was, though, if AAC encoders can beat Vorbis, and Vorbis (in my experience) always beats MP3, AAC ought to be able to beat MP3, at least some of the time.

Can you back up your claim that Vorbis consistently beats mp3? When I administered a blind listening test what I found was that, in general:
1) At lower bit rates, Ogg Vorbis files were both smaller and preferred by the listener. Both Ogg Vorbis and MP3 were distinguishable from the original.
2) At maximum quality, Ogg Vorbis files were larger and still distinguishable from the original. LAME's VBR mp3s were indisguishable.

My priorties, in order are:
1) Widely supported format
2) Indistinguishable from original
3) Save disk space

With those criteria Vorbis was clearly inferior to LAME. This was a few years ago; maybe the codec's improved.

I have seen many comparisons where Ogg Vorbis comes out ahead, but they all have defined "ahead" as quality/bit rather than absolute quality level achievable.

Well, maybe you should try lossless formats instead. FLAC (OSS + some hardware support), Wavpack (OSS + most features + better compression) and Apple Lossless (iPod support + OSS decoder). The hardware support is actually not all that essential since lossless formats are meant to be used as an archive format and you can transcode your music into any format however many times you want with no loss in quality (unlike transcoding from MP3 to Vorbis, for example)

In terms of high BR lossy compression formats, the favorite on hydrogenaudio.org is Musepack (aka. MPC, MP+), which is actually quite competitive even at 128bit as shown by the rjamorim's last multiformat listening test.

you should try the aoTuV 4.51 version:
http://rarewares.org/ogg.html

"LAME's VBR mp3s were indisguishable."
and how long did it take to encode those vbr mp3s? to make ones that are actually indistinguishable takes about 4 times as long as encoding a 256 kbps cbr mp3 (which is indistinguishable from the original to me)... and requires using the "-k" option (which no one does) to get rid of that incredibly annoying high pitched ringing...
i've actually encoded 56 kbps mp3s (try "lame -q 0 -b 56") that sound better than most vbr files i've heard.
and yes, i have done abx tests... the ringing is really there.

Edited 2005-12-15 19:54

did you try -V2 or -V2 --vbr-new and latest 3.97 b2 ? Should be quite fast. And the quality should be higher than 56 kbps CBR, not sure if you're trolling here.

And again -q0 was buggy until 3.96 or around that so you're better off using -q2 or leaving the default of -q2 or -q3.

adding --noreplaygain will speed up encoding speed even more.

Actually, (it might be my settings) I've found that Apple's iTunes MP3 sounds better at 32 kbps than Lame at 32 kbps. Granted, it could have been my sound samples but it's usually been true- excepting high bitrates.

Both sound much worse than Vorbis at the same rate, but I was trying to generate files that the average (ie, not me) person might be able to play.

Have you compared between different codecs (using ABX?)? If not, then you should state "this is just my unsubstantiated opinion".

Yes, I realized I am being very strict here, but people can make all kinds of absurd claims (eg. the article and some of the comments)

Where is your ABX result ?

From the
http://www.rjamorim.com/test/multiformat128/results.html
link the conclusion is that ATRAC3 is the worst of the lot, followed by WMA. iTunes AAC and Lame MP3 are almost tied in the middle, while Ogg Vorbis (aoTuV) and Musepack MPC are the top.

Note, however that only AAC and WMA were truely 128kbps, the others were between 3% and 6% bigger.

Also note that from earlier testing
http://www.rjamorim.com/test/aac128v2/results.html
iTunes emerged as the best AAC implementation.

Uh, I'd rather not devote an entire hard drive to my music collection when my ears can't tell the difference between the original and a Vorbis encoded file at q=4.5 anyway...

It's all about choice, and ofcourse you can choose your Vorbis files if you like, I'm not saying anything about that.

But all these tests back and forth, and making judgement calls and different types of songs which get lossy in different ways etc etc. I'm just saying, wanna play it safe, simply go FLAC and you have no need to worry at all, permanently.

puhlease Haicube

Uh, I'd rather not devote an entire hard drive to my music collection when my ears can't tell the difference between the original and a Vorbis encoded file at q=4.5 anyway...


The good thing about lossless compression is that it can be used as a master archive for whichever lossy codec popular in the future. Vorbis may sound fine, but transcoding to a different format is bad. So if the hardware players you like don't support Vorbis you have to choose between restricting yourself to the handful of players with Vorbis support or losing quality in transcoding.

A 250Gb hd can store over 600 FLAC compressed CDs and only costs between 80 and 100 bucks. That's insignificant compared to the cost of the CDs themselves.

Then why dont you just keep the CD anyway? (You have brought that...)

It's probably not your ears but your equipment. If you ever upgrade to a $500 set of speakers or a $200 pair of headphones, you'll be annoyed at your earlier self.

If you don't understand anything about the technology behind of audio encoding, then posting will only show your ignorance...

Most people criticized this article because of its flawed methodology! This article is not about fanboying OSS or Apple!!!

Such comparison between compressed signals is similar to encoding music using JPEG images. Just totally useless, academic bullshit. To compare codescs you use ABX, double blind listening tests. Otherwise you compare apples to oranges.

I think you just didn't bother to read any one of my previous posts. The whole point about lossy compression is to distort the sound as much as possible (without having you notice the distortion) to make the stream easier to compress! Thus, human perception is the ONLY way to judge an encoder!!

Therefore, the above article is unscientific because of its flawed method.

And we're not going the get one.

The idea of an universal audio codec is funny though.

The right solution is a driver/plugin for each codec or family of codecs.

On Amiga datytypes can be used for this.

The problem is that certain apps/players are using their own plugin-system instead of relying on system wide implementations.

And getting a codec to work across several different OS'es is not possible in a reasonable way (it would make an extremely large codec).

the audio format is called vorbis!