I've tried to get response for these questions for months in WMA MS Newsgroups. Only four days ago I've got an inconclusive reply (thanks again to Neil Smith), but I really would like to have more definite information. Could someone help me? See below.

While trying to figure out how much kbps should I use to have "CD quality"
in my wma library, I decided to have a look at the spectral view of several
encodings of the same music, and got surprised when seeing the wma pro 64
kbps spectrum, which seems to be "closer" to the wav at 44100 Hz than any
other encoding below wma pro 256 kbps (VBR). How can it be?

Some details:
wma at 64 kbps: frequencies beyond 13000 Hz are almost absent, except for
some "residual" sound.
wma at 128 kbps: the "cut" is near 16000 Hz.
wma at 192 kbps: the same for 19000 Hz.

wma pro at 64 kbps: the cut is slightly over 20000 Hz. (!!!!!!)
wma pro at 128 kbps: the "cut" is near 16000 Hz.
wma pro at 192 kbps: the same for 19000 Hz.

wma pro at 256 kbps (VBR): a not-so-clear cut slightly below 21000 Hz.

I'm pretty new to this audio world, but I saw a clue: the wma pro at 64 kbps
spectrum seems to have less "horizontal resolution" (sorry for the my poor
description) than the others with higher kbps. What is the trick?

By the way, I have some additional questions:

1. Windows Media Encoder has tons of encoding options. How can I use them in
WMP?
2. Which WMP encoding option offers quality equivalent to Lame3.96Final mp3 with
"--preset extreme" option (I want the most transparent sound with good
compression - no "--preset insane" need)?

Thanks in advance,

Rosivaldo.

128k to 192k if you mean transparency. I would not use WMA Pro though. Its poorly supported by most hardware. MP3 would be a better choice for your needs.



I think Pro uses SBR at low bitrates, so the spectrum will look odd.



Probably whichever VBR preset gets you around 192kbps.

Just to add a little to Mike's brief explanation: SBR is a technique for "simulating" high requencies that are not encoded due to resampling to a lower rate (in this case the audio data were probably resampled to 22.05 kHz, so there are no encoded frequensies above ~11 kHz). While this does add sonic energy above 11 kHz, it is not an accurate reproduction of the original and does not sound as good. It only works at all because our ears aren't very good at hearing frequencies this high.

I too would recommend using a different codec because WMA Pro does not seem to be catching on and there are better choices.

That depends - if you're using the audio only on Windows PCs and Windows Mobile smartphones then it's a nice no-hassles format. Otherwise, you're probably better off with HE-AAC, which utilizes the same technologies as WMA Pro (SBR, Parametric Stereo, etc), and tends to score slightly higher in listening tests on quality. Advantage is more cross-platform support (most mobile phones these days, for one), downside is that you'll have to check for compatibility a bit more (none of Apple's products support it, for one).

On "cd quality": that's in the ear of the beholder, that's why it's such an easy claim to make. 112-128 kbps seems to be the point where modern codecs (WMA Pro, Nero & iTunes AAC, LAME MP3, AoTuv Vorbis) achieve transparency in the recent listening tests. However, these tests are done by careful listening with full concentration on headphones. If you're encoding for listening on the road, in your car or on the bus with lots of background noise you'll get transparency much, much earlier (I realized that while I could pick out 64k HE-AAC pretty easily at home, I can't on my car system when I'm driving).

BTW all encoders use some kind of frequency cutoff due to the required tradeoff in all lossy encoders: "should I spend most of limited bits encoding (mostly inaudible) high frequencies or use them to reproduce the lower frequency sound better?". Hence, frequency cutoff is not a reliable criterion how good an encoder is. You can force LAME to use everything up to 22.1 KHz in a 128k encode but it will sound absolutely awful compared to one where you let LAME determine where to cut off the high frequencies.

Thank you all for the replies.

So let me condense things:

- WMAPro@64kbps uses some kind of "trick" in order to seem better than it really is.
- Any *decent* codec will be transparent at 192kbps (a lot higher than 112-128kbps).
- Therefore I can use WMAPro@192kbps with certainty that music will not degrade.
- In order to achive better portability, Lame MP3 would be better and still sound as good as WMAPro.

Now let me add some questions:

1. Lame3.96Final Docs says that ''--preset medium (...) should provide *near* transparency to most people on most music. The resulting bitrate should be in the 150-180kbps range, according to music complexity". Are they being too demanding? "Near" is not what I seek. :-(
2. As my original post says, I was aiming something equivalent to Lame "--preset extreme" (200-240kbps), which they promise to be virtually transparent in any situation (listener/music/equipment). If 192kbps is far more than enough, as you've said, isn't "--preset extreme" overkill?
3. If, instead of using WMAPro, I use VBR WMAStd set to yield 135-215kbps, may I rest sure that my music will sound transparent?

I know that my issues are somewhat silly, so some background may be useful:

1. I'd like to have music to hear on my PC, sometimes with headphones, sometimes not; or on my home theater, which plays mp3 and wma; or even in my car. I'd like to have the better sound in any situation, without having to reencode music for each case.
2. I'm not a "golden ears" guy, but appreciate music a lot. I don't want to worry about the possibility of being losing subtle nuances that good music usually has, but I'm frightened with the idea of doing double blind tests, ABX and the like. I don't care about such things when I buy a CD, and I don't want to care about this when encoding music. I'd like to be sure that my music will not degrade when encoded. This is why I'm asking for help of experienced people.

Thanks a lot for your patience and for any additional help.

Rosivaldo.

It is generally recommended to use -V2 --vbr-new with LAME.

While it is generally accepted that WMA Pro ranks right up with modern codecs in quality, lacking only in the area of compatibility across platforms, the same can not be said for WMA standard. Some testing has indicated that even at higher bit rates WMA standard has a characteristic "sound" that many people find disagreeable. I see absolutely no reason to use WMA standard under any circumstances.

Yes, yes, and yes.




You should be using LAME -V2 (what preset standard is called these days), not WMA.

All lossy codecs use 'tricks' to sound better, that's the whole point behind it. Some work very well at low bitrates, like Spectral Band Replication and Parametric Stereo - that's why these are used in WMAPro and HE-AAC. In fact, these 'tricks' are the major reasons why these modern formats sound downright amazing at bitrates as low as 32k where old codecs like MP3 and WMA Standard are reduced to a warbling sub-AM-radio unlistenable mess.

Yes for nearly all situations and music, provided that the correct presets/settings are used and an encoder that is currently still actively being developed: Microsoft's WMAPro encoder (in WMP), Apple AAC (iTunes/Quicktime), Coding Technologies AAC (in Winamp and others), FhG MP3 (used in lots of commercial apps like WMP), LAME MP3, Helix MP3 (in RealPlayer), AoTuv Vorbis.

For nearly all music and and practical applications, yes.

Correct. I see no reason to use WMAPro at >96k unless you're a major Microsoft fan. With something like 100 million iPods and 1+ billion mobile phones on this world that will play MP3 and not WMA it just does not make much sense to me.

A few years later we're now at 3.97 final, and transparency is a tad lower now. If I recall well, the december 2006 128k test already had lots of contestants complain that it was frustratingly close to transparency - and that's a test with specifically selected 'difficult' samples. The LAME switches have changed from the old "--preset xxx" to a "-Vx" format, check the hydrogenaudio.org wiki for documentation for that.

Yes, but that's why it's called 'extreme' (and 320k 'insane'). There's always some people who don't have storage space constraints but still want to use MP3.

WMA Standard is an old format and I doubt Microsoft has put much effort to optimize that encoder in the last few years now that they have WMA Pro. That said, WMA Standard will not sound terrible at this bitrate and you will probably have considerable difficulty ABXing it if you don't know what to look for. But I'm pretty sure that even the vbr mp3 encoder (licensed from Fraunhofer) in WMP11 is better (although I don't believe they've ever been tested head-to-head?).

An ABX test only need to be performed once, to convince you that you don't need a higher bitrate when you encode to lossy formats. I'm sure something around -V5 would be sufficient for you to archieve transparancy (at least most people report this as transparant). The only matter is if you are able to hear differences between the source and the encoding.

Then you should consider lossless, or you will probably find yourself reencoding in near future.

I'm sincerely amazed with the amount and quality of the information you all have given to me. Thank you very much.

Rosivaldo.

That's mainly because if the word "near" was removed, Lame devs would be immediately bashed over internet forums with claims like "they don't care bout quality", "they only care about marketing", "they are big fat liars".

The main use of the words "should", "near" and the double "most" is to shield us against bashing.

Transparent doesn't mean exactly "indistinguishable from the original"?



Only lossless formats yield transparency? After all I've read from this thread, I was convinced that mp3 can be really transparent for most purposes with right presets. Have I lost something?

Rosivaldo.

Transparent means that for a particular listener and a particular piece at a particular time, the listener is unable to distinguish the lossy encoded version from the original. It does not mean that all listeners, or all pieces for that one listener, or even that same listener and piece at a different time or in a different environment will report transparency. The best we can do is say that for many or most listeners on many or most pieces a particular encoder setting will probably be transparent.

Only lossless formats can provide guaranteed "absolutely no degradation". Transparency is when there's no audible degradation.

You just confuse "degradation of quality" and "transparency" with each other. Quality always degrades during the process of lossy encoding, no matter which quality level you choose. That's due to the encoder permanently removing major parts of the music, leaving a worse result behind. But the question is whether you're actually able to hear that this resulting file is of reduced quality, and that's where the term "transparency" comes into play. If you fail to ABX the source vs. the MP3, then the latter is considered transparent. But that doesn't change the fact that the MP3's quality was heavily suffering during the process of creating it.

For this reason encoding to MP3 should always be the last step of producing an audio file. All previous steps should be made using lossless material. For instance, decoding and modifying an MP3 in a wave editor and encoding the result once more, will cause the encoder to remove even more material from the file, possibly introducing audible flaws. The more often you lossily encode a file, the worse the result will be. That's also the reason why one should always encode from source or a lossless copy, in order to get the best possible result out of a lossy format. Despite popular belief transcoding a 128 kbit/s MP3 to 320 kbit/s won't improve the quality, but actually worsen it. You can't get back what was lost during the first encoding process.

If I can't hear the degradation, I'm more than happy. As someone said in other thread "ignorance is bliss".
:-)

Thanks for the comments.

Rosivaldo.

rfa
First of all. LAME -V 5 is transparent to me using my PC/Sound Card and My Amplifier/Speakers. It doesn't mean it will be transparent to you or it will be transparent to me with different equipment. So basically transparent means, that at current conditions you are not able to distinguish between original and compressed file.

Lossless compression is not only transparent, it's identical. It menas that any spectrum analysis or bit comparations won't be able to find any difference.

If you want to compare WMA Std. and MP3, many tests proved, that at bitrates over 128 kbps MP3 (lame VBR) will sound better. MP3 is much better supported and sound better then WMA. Why would you want to use WMA?

WMA Pro is better then MP3 but is almost unsupported. AAC and Ogg Vorbis sound probably even better then WMA Pro and are much better supported. But of course not as good as MP3.

My last point is, that comparing sound compression by spectrum analysis doesn't make any sense. Compare the sound. Any artificial analysis of lossy compressions is unusable. The only thing that matters is how good it sounds (or looks, if you compare picture/video compressions).

Spectrum analysis of AAC+ @48kbps would look much better then Ogg Vorbis 128 kbps. But I'm quite sure Ogg Vorbis@128 will be transparent to you while AAC+@48 probably wouldn't.

Mike Giacomelli
Are you sure that WMA 135-215 kbps will be transparent? I'm quite sure it won't.

Um, there is a formal test on the record between WMA Pro and HE AAC. It was discussed here some time ago, and despite some specious objections, was in fact a well-executed test between HE AAC and WMA Pro. Sorry, don't have the cite handy.

The results are quite clear, they are at the very least equal in performance, with the edge on transient signals, at least during the test, going to WMA Pro.

I can't say what, but WMA Pro does not do something like SBR. As seen in Li and Johnston's Asilomar paper from last year, WMA Pro is a waveform coding technique, although of what kind isn't specified.

My assumptions regarding audio quality were based on this listenin test. Of course the results can be completely different depending on genre, bitrate and personal preferences.

I think I read here somewhere, that there are more versions of WMA Pro. Some of them use SBR and some of them don't. I think that Microsoft will never tell how WMA Pro works, because many patented technologies would be found in their format (i.e. SBR).

Once again, I state that WMA Pro does not use SBR. Period. Please do not speculate about what you do not know. SBR is a modelling technology, the bits put into SBR, while useful perceptually, do not necessarily add to the SNR of the signal. WMA Pro uses methods that do add to the SNR of the signal, and as a result perform differently when used in scalable systems such as the one in Li and Johnston's paper at last year's Asilomar. The generalities in that paper, alone, should suffice to make the issue clear to you.

And, again, a much more recent listening test has been cited here, with much more recent encoders, etc.

This was a third-party test. I'm sorry I don't know the thread it was referred to in.

Do you mean the scalable coding paper in IEEE? Thats all I found searching, and it doesn't seem to mention what WMA Pro uses beyond an MDCT of some kind.

Notice the statements about results of 64+32 and 64+64. Try the same thing with SBR. The difference will bite you in the nose.