does anybody know if mp3's containing wavelets? coz i want to compare ogg with mp3 to detemine my format which i will use for conversion...

if it not does... i will go to OGG... else ... i don't know exactly...

thanks

Mp3 doesn't use wavelets.
Ogg also doesn't in the current version, but it probably will in the future.

Menno

Why do you need a format that is using wavelets?

To reduce pre-echo?.

From reading Frank Klemm's post I have understood that there has been no proof of wavelets working magic in any way.

For less pre-echo you can use a subband encoder such as MusePack/mpc.

Oh yes.. Wavelet is more efficient (require less bits to code) than short blocks.. If there aren't much transients in an audio clip, then wavelets isn't going to give much performance advantage overall.. However, if there is a lot of signal transients, such as clips dmb3.wav, dmb4.wav or castanets, then switching to wavelets could reduced the overall bitrates by as much as 20-30%.

As far as I know, EPAC is using a combination of MDCT and Wavelet filterbanks and they claim at 80 kbps it is producing the same quality of MP3 at 128 kbps.

wkw

Then again, so did VQF.

A high quality AAC implementation should probably do this or come very close, and they aren't using wavelets.

And FWIW, ePac never really sounded that great anyway. Their claim is probably more unsubstantiated hype than anything else.

Oh VQF is using vector-quantizations to represent the quantized spectral which is another level of lossy method whereas EPAC is using lossless huffman coding just as AAC, MP3, PAC. The developers of ASPEC, PAC, EPAC are pioneers in the field of audio coding and they are also the same people who developed MP3 and AAC!
I don't doubt their claims because they also claimed at 96kbps AAC is as good as MP3 at 128 kbps .

You could also try ATRAC3 and ATRAC 4.5 by Sony Corp which is also an excellent compression scheme. ATRAC3 do not even have short-blocks and listening tests comfirmed at 105kbps of its superiority over MP3 at 128 kbps

wkw

That depends on what bandwidth you are coding in. For Psytel AAC encoder, the castanets.wav clip cannot be coded at full bandwidth at 96kbps stereo because of the short-blocks. Of course, IS isn't implemented yet.. but that wasn't the issue..
EPAC should be able to do that at full bandwidth because of the efficiency of the wavelet filterbanks over 8 short-blocks.

Currently, there aren't any tools in AAC that resolves this problem without compromising quality.

wkw

Up to date, there is NO scientific proof that wavelet based coder is able to outperform current state-of-the-art.

By scientific proof I mean fully independent double-blind listening tests carried by at least three independent labs! Competitors must be optimized (state-of-the-art) implementations of current coding standards.

This is the way how MPEG codecs were developed, and why AAC contains exact tools that it contains now - because other tools that were tested (adaptive huffman coding, alternative quantizations, multiple window sizes, etc..) weren't showing increase of quality during the RM experiments.

I don't believe Lucent Guys - core people that developed ASPEC and PAC (JJ for examlpe) stayed in AT&T when Lucent became independent company (now Agere Systems) - and the same people claimed several times that Lucent didn't do proper listening tests, and that PAC is inferior to the state-of-the-art

There are currently a lot of universities researching into wavelets in transient compensations. Just because it isn't used in MPEG doesn't mean that it is not good.. ISO for instant did not include MIDI clips in their listening tests of MPEG2 and MPEG4 AAC.

Anyway, wavelets are used only to replaced the short-blocks not the long blocks.
EPAC for instant claims superiority only if there is a lot of transients at lower bitrates. Castanets for example could be coded at 80 - 96 kbps full bandwidth if remain in long-block mode only!

wkw

I happen to doubt their claims because I actually listened to the encoded clips.. heh.

Sorry, but ePac quality is not all that it's cracked up to be, and simply believing the claims of a company like this is foolish. How many other companies out there claim CD Quality at some odd below-128kbps bitrate, when we know this is nonsense?

Umm... I'm sorry, but I don't understand this statement.

What does MIDI have to do with audio compression at all? MIDI is only an interface for connecting and remotely controlling music devices.. why would this have any bearing on listening tests?

If you for some reason mean software synthesis driven by a MIDI file, this still has nothing to do with audio compression. Including such an example in a test of psychoacoustic encoders would be pointless. You wouldn't be testing the transparency of the encoders compression (or selective degredation), you'd be testing the quality of the software synthesis engine's sound reproduction. These are two entirely unrelated things.

Hummm... maybe he was referring to comparing MIDI with MPEG4 SA?


Edit: Yay! 1600 posts

What I meant was, encoding clips generated by MIDI or computer/machine generated audio clips with AAC encoders.. Clips like spahm.wav, fatboy.wav, awe32.wav etc-etc... Most psychoacoustic model will performed badly on these type of audio. I think you would aware that most of the problematic clips are of these types.

As for the epac samples, is there a way I could down load the test samples? Try coding dmb4, castanets with epac..

The field of audio compressions is expanding so fast that even the ISO guys could not keep up with the latest developments.. Wavelets is a very recent discovery and its used in audio coding is new. There are other new promising techniques such tone-noise-transient modelling which is not tested yet.. I knew because I used to work under some MPEG committee guys in my company.. and from my experience, even they have to struggle just to keep in touch with the latest advancement..


wkw

Atrac3: I don't know about it, but regular Atrac has mixed blocks.

Wavelets: "very recent discovery" ? I think that I don't have the same definition for the word "recent" as you.

Haar, A., :Zur Theorie der Orthogonalen Funktionen-systeme" [In german], Mathematics Analysis, Vol 69, pp. 331-371, 1910

Actually, there is atrac2, atrac3, atrac3.5, atrac4.5 atrac6.0 ....

Wavelet used in audio coding is very-very recent..

wkw

Actually there is Atrac1 and Atrac3 which are different things.

Atrac1 comes in several flavors: 1,2,3,3.5,4,4.5,4.6
This Atrac1 is for regular minidiscs

Atrac3 is a different codec, not compatible in any way with Atrac1. It is used for MD long play and computers.

About wavelets, yes, it's quite recent in audio coding. It is only researched in the audio coding field for about 10 years.

Are there any new ePAC encoders available? Last time I listened ePAC encoded files over a year ago, I wasn't impressed..
Maybe it was just that their non-transient tweaking was way off or something, I don't remember. I'd like to hear some ePAC files, both transient clips and non-transient clips.

I don't think so.. ATRAC1 comes with mixed block.. ATRAC2 & ATRAC3 comes with single block mode with Gain Control and the 2 of them are almost identical in every aspect..

The same gain-control used in MPEG4 AAC SSR Profile except that they replaced the PQF filters with QMF filters.. perhaps better impulse response and lesser aliasing between bands..

ATRAC 3.5, 4.5, 6.0, I don't know about them.. They are very recent.. I don't think they are in anyway similar to ATRAC1. There is an active team at Sony constantly improving the ATRAC algorithm.

As for epac, I have not evaluated any audio samples nor have I worked on the encoder or wavelets..

Are you saying that wavelets in Audio coding is a big flop? Some time ago, I downloaded a lot of technical papers on the used of wavelets in audio coding especially for transient compensations.


wkw

wkw. You must differentiate between Atrac(1) version 3, and ATRAC3. The latter (ATRAC3) is a lower bitrate optimized codec (MD LP).

Atrac(1) different versions are high bitrate optimized.

So when talking about Atrac 3, you must be very careful, do you mean the MD LP -ATRAC3, or the high bitrate Atrac 3.

You are right.. Even I was confused.. There are so many versions of ATRAC..
I have only seen ATRAC1, ATRAC2 and ATRAC3 but not those 4.5, 6.0 etc-etc..
I wonder if anyone has tried Real's bundled ATRAC3 encoder's. I downloaded it, have it installed but could not activate the codec components. It seemed that Real's server is down.

wkw

Do correct me if I am wrong, but I thought that something like a Discrete Wavelet Transform could be a seen as a series of band-pass filters in it's own right using discrete scales and dialates, unless of I am thinking entirley of something else . Frank Klemn might not think that wavelet's work magic in anyway, but I think you would see that Monty opinion differs slightly if you where speaking in terms of Vorbis.





yes, many universities and colleges like you said have began implementing them into experimental encoders seeking transparent results. Everyone has there own varying opinion on them. I for one think that they could provide transparent results if they where implemented correctly, however I don't have any proof to back that up. They look very promising though from an experimental standpoint though. They would only work effciently if they where implemented in hybrid form though from my understanding maybe a MDCT / DWT filterbank switch of something of that nature. MDCT/FFT for noise/tone anaylsis and the DWT for transients . I also thought that if you implemented a series of wavelet filterbank's that block switching wouldn't be needed at all? hmm...

For what it's worth, when I had originally begun talking with Monty about MPC (a subband coder), from what I understood of our discussion, he didn't seem to think that it should work as well as it apparently does.

My point is that, if this was the case, it could also be the case that wavelets would not necessarily help as much as he is thinking either.

I don't think Monty is "wrong" by any count, but I also don't think that Frank would be making a statement like this without at least some sort of testing or evidence to back up the theory. To be fair, I don't know what kind of testing Monty has done (if any), so he could have just as much evidence supporting his argument as well. However, since it is not known (at least to me), if this is the case, I'm not going to just bet on the fact that wavelets are going to make so much of a difference.

I've seen a lot of things throughout Vorbis development which were supposed to work a certain way, or solve a certain quality related problem, but which have not always had the full effect that they were supposed to (most of this backed up by listening tests). So these days, I wait and see what the final outcome sounds like rather than banking working "just because".

There's still a lot of unknowns in this field and just because something works (or for that matter doesn't work) in theory, doesn't mean it'll hold up so well in practice.

IIRC, there was a similar situation recently with one of the Vorbis stereo modes or something, and at the last minute the supposedly inferior method was switched back on because the theoretically superior one did not work as well in practice, for whatever reason.

Yes, you mean the dipole stereo rotation was switched back to the elleptical stereo line rotation. I don't know 100% about that as I don't know math extremely well, but I am willing to bet that they where polar figures related to the coupling mechanisms.



Still it's worth the experimentation like most things.