HydrogenAudio

I note that a certain pet bird has the signature line

WavPack (352/448 kbps) lossy for HQ lossy encodings and transcoding
LAME 3.90.2 (by Dibrom) --alt-preset extreme -Z --lowpass 21 or --alt-preset insane --lowpass 21 for my mp3s
MP2 384 kbps stereo for the classical music on my portable
MusePack 1.15r --insane --xlevel for archiving on CDs

Is he right in using WavPack, as opposed to MPC?

Something tells me the higher the bitrate, the more he likes it - There isn't an MP2 encoder as well tuned as Lame, and cranking the bitrate up to 384 isnt going to make the encoder more 'tuned'. And MPC insane / MP3 APX are virtually unnecessary overkills too; YMMV.

Lev:

Hahaha... you are very, but very witty pet. Why not trying it yourself? And almost everybody knows that in such high bitrates is LAME no rival for MP2, even if it's not "so tuned". Guess, what is a basis for MPC?

I don't know - what MP2 encoder do you use and what is the typical music style?

The fact that MP2 is the basis for MPC means as much as BladeEnc does for MP3 quality.

I have some opinions about the rest of that signature, but I'll keep those to myself.

First of all a lowpass at 21KHz is insane. Most modern cd players cut at 20KHz and only a few people can hear a difference with a lowpass at 16KHz. The normal lowpass the presets use should be safe enough.

Secondly. Do you consider wavpacks lossy mode to be of a higher quality than mpc? I don't think even the author of wavpack would agree.


Last but not least I would like to see an mp2 encoder that can outperform LAME. Though it shouldn't be impossble I certainly have not heard about such an encoder.
To the claim that mp2 should be good because mpc is I can only say that that is really a flawed argument. Is xing good because LAME is relatively good? MPC is way beyond what mp2 ever was/is.


edit: ok, after search this board it seems that wavplack lossy mode is actually quite good at that high bitrate and a competitor to mpc (though mpc would use a lower bitrate).

Very original, I don't ever recall some one starting a thread over a member's signature. It will be interesting to see the results MPC+WPL ABX testing.  It really annoys me to see: LAME 3.90.2 (by Dibrom) should be 3.90.2 compiled by Dibrom or credit all the lame developers.

HA History X.

Edit: Interesting, Garf edited his post without a "This post has been edited"

The normal lowpass the presets use should be safe enough.

--ARE safe.  Its not the lack of High Freq. which lets MP3 down, its the constraints of the format, and resulting smear / pre-echo (which will worsen with upped lowpass).  [Luckily, I dont find these annoying]

Where can I get a Wavpack Encoder / Decoder / Plugin?  (Yes, I know, UTFS, sorry)

Whats the general consensus for quality?
If this is the standings at the moment:

1. MPC
2. AAC [span style='font-size:8pt;line-height:100%'](dont understand the differences between all the different implementations / versions / modes, so sorry if its unfeasible just to write 'AAC')[/span]
3. OGG
4. MP3

Where would WavPack lossy show up?  1.5, 2.5, 3.5 or 5

[span style='font-size:8pt;line-height:100%']Edit: I've just realised this is liable to generate a huge number of responses from people defending their preferred formats.  Can we just stick to the positioning of Wavpack, please
Thanks[/span]

Edit: Interesting, Garf edited his post without a "This post has been edited"

That is default for all moderators I think. There's a checkbox I have to explicitly tick off if I want the 'has been edited' stuff to show up, which I generally won't think about when I fix a small typo or clarification immediately after posting.

It will be interesting to see the results MPC+WPL ABX testing.

--Will be provided, probably not over the extreme drinking binge known as Easter

First of all a lowpass at 21KHz is insane. Most modern cd players cut at 20KHz and only a few people can hear a difference with a lowpass at 16KHz. The normal lowpass the presets use should be safe enough.

Secondly. Do you consider wavpacks lossy mode to be of a higher quality than mpc? I don't think even the author of wavpack would agree.


Last but not least I would like to see an mp2 encoder that can outperform LAME. Though it shouldn't be impossble I certainly have not heard about such an encoder.
To the claim that mp2 should be good because mpc is I can only say that that is really a flawed argument. Is xing good because LAME is relatively good? MPC is way beyond what mp2 ever was/is.


edit: ok, after search this board it seems that wavplack lossy mode is actually quite good at that high bitrate and a competitor to mpc (though mpc would use a lower bitrate).

Most modern crappy-shitty CD players maybe cut at 20 kHz, but not a decent one. Every CD player worth its name has frequency range at least 20-22050 Hz. When is this insane, why AAC/MPC encode at full bandwidth? (Not the standard settings). Ogg also encodes far beyond 21 kHz  So it's no insanity.

WavPack lossy mode at this bitrates is for me definitely better than MPC. Although it doesn't mean MPC is wrong or of insufficient quality. My reasons were stated here http://www.hydrogenaudio.org/forums/index....&f=1&t=8280&hl= (http://www.hydrogenaudio.org/forums/index.php?act=ST&f=1&t=8280&hl=)

As for the author of WavPack... I think no developer of any codec would express to the another codec in terms of quality. So we should leave it on their decision. But I personally believe the WavPack author knows his codec is of very high quality, no doubt about it.

MP2 was designed for high bitrates, whereas MP3 not. That's why (but not due to this only fact) it sounds good at these bitrates. This is from one of my PMs I got:

MP2 really rules above ~250kbps, indeed. About the same thing as musepack ~180kbps.

If the author allows me it, I will reveal his name.

It really annoys me to see: LAME 3.90.2 (by Dibrom) should be 3.90.2 compiled by Dibrom or credit all the lame developers.

Excuse me, sir, I really didn't want to offend your feelings...  Already corrected to your satisfaction, I hope so...

Most modern crappy-shitty CD players maybe cut at 20 kHz, but not a decent one. Every CD player worth its name has frequency range at least 20-22050 Hz. When is this insane, why AAC/MPC encode at full bandwidth? (Not the standard settings). Ogg also encodes far beyond 21 kHz  So it's no insanity.

--bah..  You ^might^ be able to hear a 22khz tone, but unless it was just that tone at high intensity and nothing else, you wouldnt hear it in music.  No chance. 

MP2 was designed for high bitrates, whereas MP3 not. That's why (but not due to this only fact) it sounds good at these bitrates. This is from one of my PMs I got:

Quote

MP2 really rules above ~250kbps, indeed. About the same thing as musepack ~180kbps.

If the author allows me it, I will reveal his name.

--(http://gene.wins.uva.nl/~mwvantol/dogbirt.gif)

[span style='font-size:8pt;line-height:100%']Edit: Sorry, thats over the top, but its well natured, done with a smile and not intending to offend (and its funny)[/span]

Most modern crappy-shitty CD players maybe cut at 20 kHz, but not a decent one. Every CD player worth its name has frequency range at least 20-22050 Hz.

The issue is far more complicated than this, in a so much that I wouldn't dare to do a statement who is right here.

When is this insane, why AAC/MPC encode at full bandwidth? (Not the standard settings). Ogg also encodes far beyond 21 kHz   So it's no insanity.

Last I looked, MPC at standard level was lowpassing at 20kHz. Ogg at standard level is below that even. The most aggressive modes are called 'insane' and 'braindead' for a reason.

WavPack lossy mode at this bitrates is for me definitely better than MPC. Although it doesn't mean MPC is wrong or of insufficient quality. My reasons were stated here http://www.hydrogenaudio.org/forums/index....&f=1&t=8280&hl= (http://www.hydrogenaudio.org/forums/index.php?act=ST&f=1&t=8280&hl=)

What I read from that is that WavPack is better for you because it gives you the warm fuzzy feeling (even though you admit MPC is tranparent for you).

As for the author of WavPack... I think no developer of any codec would express to the another codec in terms of quality.

What??

MP2 was designed for high bitrates, whereas MP3 not.

This is true. It in no way invalidates the points that were made.

That's why (but not due to this only fact) it sounds good at these bitrates.

I've asked you about the encoder and settings you use. You didn't respond. Please do.

Specifically:

Once the quality exceeds what can be measured in ABX tests, I suspect that most of the people at HA would say that all codecs are the same, but I don't believe that. There is no doubt that WavPack's lossy mode provides better quality than other lossy codecs above some bitrate because it provides a straight line to lossless encoding which the others do not.

There are no logical grounds for either statement.

--bah..  You ^might^ be able to hear a 22khz tone, but unless it was just that tone at high intensity and nothing else, you wouldnt hear it in music.  No chance.

Dear Lev, it really doesn't matter, if you can hear it or not. Or do you think all the developers of these codecs are stupids who don't know what they are doing? Also all the CD players' manufacturers? Or those who record and mix the CDs? Why was SACD or DVD-A developed? Just for the fun of it? This matter was discussed here many a time, I guess. Stop repeat BS you read here and there on the HA and really learn something about the matter. It would really help you, believe me.

Stop repeat BS you read here and there on the HA

--  It aint all BS on here!

I'll provide some FLACs for you to ABX next week ref: lowpass @ 20khz

[span style='font-size:8pt;line-height:100%']Edit:  Noticed this

Or do you think all the developers of these codecs are stupids who don't know what they are doing?

... Ummm, I think of them as Gods to be quite honest, thats why they lowpass in the first place![/span]

Dear Lev, it really doesn't matter, if you can hear it or not. Or do you think all the developers of these codecs are stupids who don't know what they are doing?

See my statement about default MPC and Vorbis lowpass. I did part of the work in tuning Vorbis for high bitrates and I can tell you that the sole reason the highpass is gone in higher quality modes is to prevent people like you from bitching that it lowpasses (since the bitrate cost is negligible, a tradeoff I gladly make).

If you load in a Vorbis file, even encoded at high bitrates, you'll see it'll almost never goes over +- 19kHz.

Also all the CD players' manufacturers? Or those who record and mix the CDs?

Some CD's are pre-lowpased at 20kHz. Don't find them much anymore, but then again, not lowpassing fits in with the trend to compress all dynamics out of them I guess.

Why was SACD or DVD-A developed?

To push DRM.

This matter was discussed here many a time, I guess. Stop repeat BS you read here and there on the HA and really learn something about the matter. It would really help you, believe me.
[/qoute]

I couldn't have said it better, why don't you do this?

Garf:

Last I looked, MPC at standard level was lowpassing at 20kHz. Ogg at standard level is below that even. The most aggressive modes are called 'insane' and 'braindead' for a reason.

If you read carefully, I wrote (Not in the standard settings).

What I read from that is that WavPack is better for you because it gives you the warm fuzzy feeling (even though you admit MPC is tranparent for you).

That's your interpretation. I never said anything like that. I mentioned spatial resolution, no "warm fuzzy feeling".

I've asked you about the encoder and settings you use. You didn't respond. Please do.

You can find it on Roberto's site. Everything else is in my sig.

If you load in a Vorbis file, even encoded at high bitrates, you'll see it'll almost never goes over +- 19kHz.

I tested a lot with Vorbis 1.0 at -q4 and the frequency response war up to 21,3 kHz...

Some CD's are pre-lowpased at 20kHz. Don't find them much anymore, but then again, not lowpassing fits in with the trend to compress all dynamics out of them I guess.

A lot more isn't. No reason for doing it.

To push DRM.

I simply don't believe this is the only reason. It gonna get cracked as everything else.

I couldn't have said it better, why don't you do this?

Just because you are one of the few people here I respect for their work, I won't react to this...

If you read carefully, I wrote (Not in the standard settings).

MPC is designed to be transparent at the standard setting. It lowpasses at 20kHz. What does that tell you? I expained you the reasons why Vorbis doesn't. APS also has a lowpass (though there are other pressing reasons in MP3).

That's your interpretation. I never said anything like that. I mentioned spatial resolution, no "warm fuzzy feeling".

If you cannot ABX a difference, all that's left is placebo. Hence a 'warm fuzzy feeling' is the only difference, except that you're calling it 'improved spacial resolution.'

I simply don't believe this is the only reason. It gonna get cracked as everything else.

I hope it gets cracked. But they put CSS on DVD's too. But you are right, another good reason is that they can 'sell people new expensive equipment'. (Those are Frank Klemms words)

Just because you are one of the few people here I respect for their work, I won't react to this...

I've happily flamed  Dibrom, JohnV, Monty and David Robinson in the past, so please don't spare me your sound reasoned out arguments, I don't deserve it.

I tested a lot with Vorbis 1.0 at -q4 and the frequency response war up to 21,3 kHz...

Vorbis 1.0 lowpasses at 20.5kHz at quality 4, and the default quality is 3 IIRC.

(I'm guessing the higher freq you see is actually a compression artifact, or a bad spectrum analyzer)

Why was SACD or DVD-A developed?

Companies are in business to make money

[span style='font-size:8pt;line-height:100%']Edit:Actually, I dunno.  I need to read up[/span]

I may be wrong but doesn't SACD & DVD-A allow easier mastering & not allow clipping & compression? Error correction would be much better than current CDDA also surround sound support but obviously they are not compelling enough for general users to switch.

And better spatial restitution (multichannel)

I may be wrong but doesn't SACD & DVD-A allow easier mastering & not allow clipping & compression?

It's not about capabilities - CDDA has those as well, at best it's slightly harder for stupid mastering engineers to shoot themselves in the foot. The only 'advantage' is that early on there will be incentive to properly master SACD/DVDA (and make even worse CDDA) to demonstrate the 'advantages' of the new formats.

You can make shit hot SACD and DVDA without dynamics as well, I'm sure it wont take long before they are commonplace as well.

Error correction would be much better than current CDDA

Is it of any significance? Many CDDA's nowadays have crippled error correction for 'copy protection'.

surround sound support

Ok, maybe a small advantage

Quote

Error correction would be much better than current CDDA

Is it of any significance? Many CDDA's nowadays have crippled error correction for 'copy protection'.

True, but with DRM integrated into the disc format, we can have our error correction back.    :duck n' run:

But you are right, another good reason is that they can 'sell people new expensive equipment'. (Those are Frank Klemms words)

And don't forget about possibility to once again sell people all those albums they already bought (already more than once in some cases). It's easy once you make them belive that new format sounds much better.

I wonder how long we'll wait for format that "breaks through some significiant limitations" of DVD-A/SACD   

Where can I get a Wavpack Encoder / Decoder / Plugin?  (Yes, I know, UTFS, sorry)

Normally, WavPack is available at www.wavpack.com (http://www.wavpack.com) but I am currently between hosting providers and (surprise!) it did not go exactly as planned. It should be resolved (no pun intended  ) in a few hours.

BTW, for the purpose of evalulating WavPack's lossy quality it is important to use the "high" quality mode (which is not the default). This is an option to the command-line program (-h) and also available in Speek's frontend and Spoon's plugin for dBpowerAMP.

Where can I get a Wavpack Encoder / Decoder / Plugin?  (Yes, I know, UTFS, sorry)

At my 'Other Stuff' Mirror 1.

Anyone else sick of of sigs with lists of favorite codecs/commandlines etc.?

Back to the original question... how good is Wavpacks lossy quality in comparison to the other codecs used here?  I wouldn't stand a chance of ABX'ing myself, it doesn't go as low as 60kbs afaik  Someone with ears please do tell!

@floyd
I know what you are saying, but I find having the sigs useful to see what others are trying. If it wasn't for these sigs, I wouldn't have tried Vorbis believe it or not... Also thanks to budgie, I was also enticed to try Wavpack. 

@everyone else
I think every forum needs a budgie. Otherwise people tend to get into the forum's main groove, and start answering every newbie inquiry with the standard answer. You might not agree with budgie and sometimes budgie posts opinions without backing it up, but the posts make you think outside the square you live in. 

@mac and others who are interested.
Curiosity killed this cat, so I dug up the latest wavpack and did some WinABX tests. I don't think of myself as having golden ears (I love Minidisc afterall!), but I can often pick out Ogg Vorbis q6, LAME aps and very occasionally fluke ABXing MPC q5 with my normal music, 70-90's rock/pop stuff. I've never tried the problem samples.

I tried the first 10 seconds of Mandela Day - Simple Minds from their Street Fighting Years CD. With LAME aps and Ogg Vorbis 1.0 q6 I can get this virtually everytime. Nothing major, just minor swirliness in some of the piano chords and something in the percussion not being quite as sharp., but it's very close. MPC q5 I'm usually just guessing!

With Wavpack 256 kbit lossy, I hit 15/15. Wavpack 320 kbit lossy, I hit 12/15. Wavpack 352 kbit lossy, I hit 8/15 initially and then got 10/15. The thing is though, I'm not hearing significant problems at any of these rates, the main thing I'm picking up is background hiss. It's slightly more pronounced in the Wavpack version, and it get's louder as you go down in bit rate. 

I tried 128 kbit without ABXing, and in this small segment, there were no killer artefacts that I could detect without concentrating. In comparison MP3 and Vorbis at these bitrates introduce an obvious wateryness in the piano stabs, and the percussion starts to smear.

What does this all mean? Maybe there is another option out there for archiving your CDs. Go Wavpack hybrid at whatever bitrate you can afford/put up with, keep the .wv file on your harddisk for listening, the odd transcode (in my case record to MD) and keep the .wvc (correction) files on CDRs. Then you have a set of reasonable quality lossy files for everyday use, and the same music in lossless when you need to do a serious transcode, or want the original wav file...

As I said, I don't have golden ears, and would love to read the opinions of some of the other members out there...

I can't easily post my samples for the time being, but perhaps some else can try some of their own music and post their thoughts...

Den

Editted to change second lossy to lossless in third last paragraph (19/4/2003)

Me again.

I've been doing some further testing, and again, I'm ABXing Wavpack lossy@352 consistently, but not from artifacts, just subtle differences in background hiss. Just to make sure I wasn't doing something weird, I tried to ABX Wavpack lossless, and as you would expect, I can't as it is identical. I just wanted to make sure that there wasn't some weird noise being added as I encoded. 

Also in my last post, I commented regarding an informal listen at 128kbit being free of obvious artifacts. Ignore that statement, as I'm finding that when I use Speek's front end, I can't seem to compress anything below ~230 kbits. If I put in anything less than ~230, I still get a 230 kbit file with the stuff I'm testing. Everything above that works OK though, and I'm still impressed with the quality. 

230kbit is still quite good, but the subtle background noise becomes more obvious in quiet passages. Not distracting, but you notice it because it's not normally there with the same passage as a musepack, or high bitrate mp3/vorbis file. I like how Wavpack does not add rubbish to the music as such, it just adds more hiss-like background noise, at least to my ears.

I'm going to dabble with this "codec" a bit more, and perhaps report my opinion later. I think I may have found a cool compromise for me, keeping the lossy files on my PC for everyday transcoding to Minidisc through foobar, and keeping the correction files backed up so I can go truly lossless when I choose. From here I going to rip and encode a bunch of my everyday listening stuff, and see how it goes.

Den.

Thanks for the interesting reports, den.

I thank you too. I find listening tests with WavPack lossy mode very welcome, since it's an unique form of compression (comparing to the others discussed here), and there's a lack of comprehensive tests about it.

Cheers.

OK, some follow up comments about this rather unique form of audio compression.

My main interest here was to see if Wavpack lossy could provide a better compressed lossy source for further transcoding. Because I use Minidisc with its NetMD capability everything I transfer ends up in ATRAC3. I also have other devices that use other formats, and the Sony software is pretty ordinary for archiving and maintaining a music library, so I have been archiving everything as higher end vorbis, musepack and mp3 and then transcoding these as required into ATRAC. Lossless would be the ultimate for me, but I just don't have that much space for all my cds.

I took a favourite track of mine, Blue Monday - New Order (the original 12") and did the following:
1. Original wav into ATRAC3 via foobar and Sony OpenMG for replaygain.
2. Original wav into Wavpack lossy @ 320 kbits then to replaygain'd wav via foobar, then ATRAC3.
3. Original wav into mpc q6 (v1.15r) then replaygain'd wav via foobar, then ATRAC3.
4. Original wav into ogg q6 (v 1.0) then replaygain'd wav via foobar, then ATRAC3.
5. Original wav into mp3 aps via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.
6. Original wav into ogg q3 then replaygain'd wav via foobar, then ATRAC3.
7. Original wav into mp3 128 CBR via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.

I then transferred all of the above onto Minidisc and sat through them all, listening to certain sections, but particularly the classic drum machine intro, and snares at the end of the intro section, and the jet flyover in the middle.

I could wax lyrical about all sorts of little artifacts that popped up in different sections, but in summary, every track had some artifacts, thanks to the wonder of ATRAC3 encoding. (ATRAC3@132 kbit is kinda like mp3@128 kbit, the average listener will say, "wow, totally transparent, CD quality!!!", but some practice and intimate knowledge of the original recording will reveal minor artifacts.  )

What stood out though was the similarity between wavpack lossy and the original wav when listened to as ATRAC3 encodes on MD. They were effectively identical. As mentioned in my previous posts, there was a very slight increase in overall background noise as faint hiss, but it was very quiet, and this is an analogue recording anyway. The other minor ATRAC3 introduced artifacts were identical between wav and wavpack, despite the wavpack being a transcode.

All of the others showed varying degrees of additional artifacts introduced by transcoding. In terms of my preference quality wise, mpc q6 was the next best, followed closely by mp3 aps and vorbis q6. The mpc added some subtle clicks and pops in the drum machine intro, while the last two introduced additional smearing in the drum machine intro and snares. (I'm guessing the mp3 and vorbis are adding some extra pre-echo to ATRAC3's pre-echo.) Then the low bit rate mp3 and vorbis really showed what transcoding can do to your music. THe drum machine beats started to run together, with pre and post echo. I'm sure to many of my friends they would still be very acceptable for everyday listening, but not so good when you have heard better.

It would appear to me now, that my best choice is wavpack lossy @ 320 kbits, (perhaps with the correction files archived somewhere as well) seeing as I regularlly transcode to MD, and sometimes mp3. For others who transcode, there could be something here for you as well.

Den.

1. Original wav into ATRAC3 via foobar and Sony OpenMG for replaygain.
2. Original wav into Wavpack lossy @ 320 kbits then to replaygain'd wav via foobar, then ATRAC3.
3. Original wav into mpc q6 (v1.15r) then replaygain'd wav via foobar, then ATRAC3.
4. Original wav into ogg q6 (v 1.0) then replaygain'd wav via foobar, then ATRAC3.
5. Original wav into mp3 aps via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.
6. Original wav into ogg q3 then replaygain'd wav via foobar, then ATRAC3.
7. Original wav into mp3 128 CBR via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.

Apples to oranges. Congratulations on showing that Wavpack lossy @320kbps can beat everyone else at ~200kbps (Except for 6 and 7 which are ~112 and 128kbps, of course).

I did a little test to see how wavpack hybride mode would perform compared to a quick-n-dirty musepack+monkey's audio combo.

The original .wav file was 54 072 524 bytes in size ("The Sails of Charon" performed by Yngwie Malmsteen).

1. I encoded it to .mpc using mppenc 1.14 with --quality 8 --xlevel.

2. Then, I decoded the mpc file to .wav

3. Using mix-paste with "invert" checked in Cooledit, I subtracted the decoded wav from the original, and then encoded the resulting "difference" file with monkey's audio using "extra-high" compression. Note that for some reason mix-pasting difference file over decoded mpc gave some very rare last-bit errors (72 in total for that file).

I compared the file sizes with wavpack 3.97 hybrid mode using -b352 -c -h as well as the original file compressed with monkey's audio "extra high"..

The results were:

.mpc file size - 10206936 bytes (266 kbit/s)
difference.ape -  30494008 bytes
total mpc+ape size - 40700944 bytes

.wv file size - 13491318 bytes (352 kbit/s, as expected)
.wvc file size - 25959505 bytes
total wv+wvc size - 39450823 bytes.

Original compressed with monkey's audio was 38014984 bytes,
and the total size of original.ape and original.mpc was 48221920 bytes.

Thus, encoding to mpc and then compressing the difference with monkey's audio makes certain sense, as it offers a 15% improvement over keeping both the lossless encoding of the original and the mpc file.

Compared to this "mpc+ape hybride encoding" , wavpack hybride mode at 352 kbps offers an additional 3% decrease in size, at the expence of a larger lossy file, as well as (according to den) decreased transparency.

I would believe Den used settings where codecs become transparent, rather than caring about bitrate consistency.

Quote

1. Original wav into ATRAC3 via foobar and Sony OpenMG for replaygain.
2. Original wav into Wavpack lossy @ 320 kbits then to replaygain'd wav via foobar, then ATRAC3.
3. Original wav into mpc q6 (v1.15r) then replaygain'd wav via foobar, then ATRAC3.
4. Original wav into ogg q6 (v 1.0) then replaygain'd wav via foobar, then ATRAC3.
5. Original wav into mp3 aps via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.
6. Original wav into ogg q3 then replaygain'd wav via foobar, then ATRAC3.
7. Original wav into mp3 128 CBR via LAME 3.90.2 then replaygain'd wav via foobar then ATRAC3.

Apples to oranges. Congratulations on showing that Wavpack lossy @320kbps can beat everyone else at ~200kbps (Except for 6 and 7 which are ~112 and 128kbps, of course).

hmm agreed... at least I would have been interested how MPC insane/braindead compares to wavpck 320k for the sake of fairness (not that I would use/recommend  such high MPC settings, as stated that -q6 is already very close).

@moneo: have you tried doing the inverse with APE+MPC to see if the result is really the same as the original wav?

I personally would still prefer using Wavpack hybrid. First because it's easier , second because I can play back losslessly if the correction file is in the same directory. I can't do that with APE+MPC.

have you tried doing the inverse with APE+MPC to see if the result is really the same as the original wav?

Yes.

There were 72 last significant bit errors. I think that it was because of the difference between two files going over 16 bit treshold at these positions.

I also agree that the whole process is very clumsy.

rjamorim
According to den, mpc at --quality 5 sounded transparent to him, and wavpack's lossy encoding at 352 kbps didn't. That's what I mean by "decreased transparency".  I didn't perform any listening tests myself.

Edit: using a simple utility that subtracts wavs and gives bit-identical results, diff.ape increases by only 24 bytes, so the results are essentially the same.

rjamorim
According to den, mpc at --quality 5 sounded transparent to him, and wavpack's lossy encoding at 352 kbps didn't. That's what I mean by "decreased transparency".  I didn't perform any listening tests myself.

That's true.

BTW: @Den: are you using WavPack lossy at high quality mode? (-h)

Just to clarify a few things -

Yes, I chose the mpc q6, vorbis q6, lame aps based on what either I have tended to find is transparent, or what others (including myself) have used when transcoding, either to MD, or to mp3s for their friends.    On many other threads in this forum, members have mentioned transcoding from mpc q6/ vorbis q6 to mp3 for sharing, car players, portables, etc. Transcoding also seems to be budgies main use for wavpack lossy.

Yes I am using the high quality option in wavpack. Haven't tried it without.

Apples to oranges, I agree. I am going to repeat the test sometime to with more closely matching bitrates. After all mpc, vorbis and mp3 @ 320 kbits could be closer to the wave and wavpack results in terms of less introduced artifacts from transcoding, but I suspect that won't be the case. Wavpack doesn't apply any signficant psychoacoustic modelling AFAIK, it just applies its normal lossless compression, and then starts throwing away "the least significant bits" to get to the specified bitrate. (Would love a clarification of what that means from the developer  ) The others apply specific models, and will all introduce some artifacts at any bitrate. It's a case of whether they can be heard or not. As has been posted on here many times, if you take a problem sample, even with mpc, and keep upping the bitrate, some artifacts do not go away, they just become less obvious, even at q10. I suspect something similar will happen with mpc in this case, and these will be "amplified" by the transcode to ATRAC3 but I need to test.

Lastly, I included the last two for comparison, and also because I intend to use these results to educate some of my brothers in the MD 'hood. They all like to talk about how 128 kbit from Kazaa rulz, and low bitrate oggs are perfect on their MDs. 

After all, if I can get a similar, or better result for my particular needs (transcoding to MD) with a higher bitrate mpc, vorbis or mp3, which is still below 320 kbit, this lossy wavpack option has less appeal, but on the other hand, these others don't offer a lossless mode if I retain the correction file elsewhere. 

I have a heap of other tests I wish to perform, and as I progress, I'll post the results.

Den.

OK, just snuck in two more test tracks, the same song (Blue Monday - New Order) encoded in mpc@q10 (beyond braindead!) and LAME 3.90.2 MP3@insane. The mp3 was obviously at 320 kbits, and the mpc was just under. (~315 IIRC, I don't have the files here on this PC).

As I suspected, these could be still be distinguished from the wav and wavpack lossy examples when transcoded into ATRAC3@132 kbit and played back on Minidisc. The mpc was still very good, but there were discernible small pops and squeaks between the beats in the drum machine intro. Once things got going in the middle of the song, the rest of it was very close to the wav/wavpack samples.

The LAME track was also signficantly better than the previous mp3 and vorbis tracks, but still slightly behind mpc@q6 tested previously. There were some extra noises and slight smearing in the intro and the highhat and snare sounds, but it was significantly better than the vorbis q6 and LAME aps tracks. When the song gets into the thick of it, the bass guitar "melody" and percussion sections were also not quite as "punchy" as the wav/wavpack/mpc examples. I'm guessing that most of this is due to the double pre-echo on pre-echo from both mp3/vorbis and ATRAC3. When listened in isolation, these effects are hardly noticeable, but they become more obvious listening to the tracks back to back.

I also tried to blind ABX test my tracks. It isn't easy to ABX test with Minidisc, but I got my wife to shuffle tracks around without telling me the selection, and when doing this, I could consistently group the wave/wavpack lossy together, mpc q10/q6/LAME api together, LAME aps/vorbis q6 together and mp3 128kbit/vorbis q3 together, but not readily distingush within each of these groupings, every single time.

This is telling me that Wavpack lossy@320 is probably the best for my minidisc transcoding needs, so it might be time for me to update my signature! 

I still want to try some straight comparisons between Wavpack lossy, Vorbis, Musepack and LAME without the transcode to Minidisc at a later date, but would be interested to hear other points of view on Wavpack lossy's usefulness as a lossy codec.

Hope this has been useful.

Den.

Great! Truly great. Thanks a lot for your tests, Den. They are definitely very helpful.

This is telling me that Wavpack lossy@320 is probably the best for my minidisc transcoding needs, so it might be time for me to update my signature!

Indeed.

@Moneo

Sorry for not respondng earlier. I probably didn't make myself clear. As far as transparency was concerned, Wavpack lossy at 352 and 320 kbit is effectively transparent to me. It sounds the same in terms of the actual music with no audible artifacts. What I was ABXing every time was a slight increase in background hiss. At these bitrates, the hiss was very slight, but with the volume cranked up, it was there. I've gotta say that to pick it, I had the volume considerably higher than what was actually comfortable for listening to the music itself and the drum machine intro was hurting my ears.    Not exactly normal listening conditions.

Your hybrid MPC+Monkey's is an interesting point, and I've heard of others who package Ogg vorbis and FLAC files together. As Rjamorim has said though, the Wavpack is certainly easy to convert backwards and forwards. If you happen to have the .wv file by itself, it is lossy, but if you happen to copy the .wc file across into the same folder, instant lossless instead. Very convenient.

My last tests also showed that even at MPC q10 transcoded to ATRAC3 was not transparent against original wav direct to ATRAC3, but Wavpack @ 320 kbit was if you were not listening for the slight increase in background hiss.

Thanks for your comments, and also others who reminded me of the importance of comparable bit rates.

Den.

Wavpack doesn't apply any signficant psychoacoustic modelling AFAIK, it just applies its normal lossless compression, and then starts throwing away "the least significant bits" to get to the specified bitrate. (Would love a clarification of what that means from the developer  )

First of all, Den, thanks a lot for testing out WavPack and taking the time to describe your results. I appreciate it.

The easiest way to describe the way the lossy mode works is to compare it to regular lossless. Anyone who has messed around with lossless audio compression has noticed that the lower the sound level of the track, the better the compression. In fact, just by reducing the level of an audio sample it's possible to get just about any arbitrarily low bitrate. This could be done manually by normalizing a track down to (for example) 10%, then losslessly compressing it, then cranking it back up after decompression.  This would work fine, but the problem is that the quantization noise added would be constant throughout the track and so it would probably be audible during the quiet portions and you would still be wasting resolution during the loud parts. The whole problem here is that the ear works logarithmically and PCM is just too, well, linear!

The way WavPack lossy works is to modulate the added noise continuously so that it is [hopefully] always masked by the music to the same degree. This means modulating slowly enough that it doesn't spike during transients (or when the predictor is adjusting) but quickly enough that it doesn't become audible during quiet parts of the music.

If you want a more detailed description of that's going on there's a few paragraphs about it at the beginning of the words4.c module of the source code also.

As far as transparency was concerned, Wavpack lossy at 352 and 320 kbit is effectively transparent to me. It sounds the same in terms of the actual music with no audible artifacts. What I was ABXing every time was a slight increase in background hiss. At these bitrates, the hiss was very slight, but with the volume cranked up, it was there.

I don't mean to spoil your party, but background hiss (slight or not) is an artifact. 

hans-jürgen wrote

I don't mean to spoil your party, but background hiss (slight or not) is an artifact.

Yep. you're absolutely right. Just to clarify, when I used the word artifact above, I am referring to the audible, additional sounds that are being added and alterations to the actual music itself. These include pops, clicks, smeared attacks, fluttery decays during sustained notes, ringing and others. All of these were detectable to some degree in all of the mpc, vorbis and mp3 transcoded ATRAC files at all tested bit rates. I could not detect any of the above characteristics over those introduced by ATRAC on top of the original wav file when I transcoded from Wavpack lossy @ 320 however.

Instead Wavpack introduced this slight hiss, and yes, it is technically an artifact. I should have expressed this differently. What I meant was that to my ears, it was preferable to have a slight, but barely perceptible increase in background hiss, over the quite distracting additional noises that were being introduced by the other codecs. Some mp3 portables I have listened to in the past added more hiss than Wavpack during playback due to their cheap circuits. My MD at least seems to be based on high quality clean sounding electronics.  Which makes the hiss a little easier to detect.

Some people might prefer the extra bits and pieces added to the music, others might prefer the hiss. I prefer the hiss.

Thanks for the advice regarding my terminology.    It's worth getting it right.

Den.

I have wanted to test this out myself, but sadly I managed to lose around half of my files while formatting my computer

I hear that the Winamp (or whatever) plugin can automatically detect the correction file and play back lossless files to you if the lossy & correction are in the same directory...  how much strain does this add to the decoding process, it's taking 2 streams and adding them together on the fly to my imagination, does this provide a significant increase in decode time/difficulty?

I have to say, I'm interested by the sound of this encoder

Mac,

I can't comment on the Winamp plug-in as I have only tried foobar which plays Wavpack back natively.

On my Duron 700 desktop system though, foobar plays back the lossy, or the lossy + correction file without any difficulty. I haven't checked CPU usage, but it doesn't choke, and you can seek through the files readily, although seeking is not quite as quick to respond as the other lossy formats, like mp3, mpc.

As you say, when foobar only sees the .wv file in the directory, you get lossy. If the .wc file happens to be present as well, you get lossless. Very convenient, and very easy to switch between lossy to lossless by copying or removing the .wc file.  B)

I haven't noticed any significant slowdown when using the correction file to get lossless, but I haven't tested it thoroughly.

Den.

Very convenient, and very easy to switch between lossy to lossless by copying or removing the .wc file.   B)

Actually, you can switch on/off using the correction file in the Winamp plugin's settings. No need to delete/rename it.

I don't know if that option is available in Foobar.

@ rjamorim

1. As far as I can tell, you can't turn it off or on in foobar.

2. I followed your previous comment and revised my signature.  B)

@ others
Have completed some other follow up tests with a wide range of ambient and punchy pop/rock, focusing only on mpc beyond braindead, LAME insane and Ogg Vorbis q10. Found artifacts in all cases to varying degrees if I listen hard enough when they are transcoded into ATRAC3 (Minidisc). LAME and Ogg Vorbis were considerable worse than MPC in terms of introduced artifacts with transcoding, but all three were not too bad.

I'm not getting any extra individual noises or changes in the music with Wavpack lossy @320 when transcoding, apart from the slight increase in background hiss at very high listening volumes in a quiet room, so now its Wavpack lossy for me. Thanks budgie for making me aware of this possibility, and thanks to bryant for making this kick-arse lossy/lossless combination available.

@bryant
If you are ever looking for someone to test new versions, feel free to drop me a line. Also, has anyone ever compiled a Linux version from your source code? I'm currently considering setting up a Linux box for audio encoding, storage and serving purposes.

Den.

2. I followed your previous comment and revised my signature.  B)

Great.

Have completed some other follow up tests with a wide range of ambient and punchy pop/rock, focusing only on mpc beyond braindead, LAME insane and Ogg Vorbis q10. Found artifacts in all cases to varying degrees if I listen hard enough when they are transcoded into ATRAC3 (Minidisc). LAME and Ogg Vorbis were considerable worse than MPC in terms of introduced artifacts with transcoding, but all three were not too bad.

Excellent. That's a proof of the theory that subband codecs (MPC, MP2) behave better than transform codecs (MP3, AAC, Vorbis) when being transcoded.

Dunno about others' opinions, but I find these findings extremely significative. Before, people only said "theoretically, transform is worse than subband for transcoding". Now, we can be almost completely sure that theory applies to practice.

has anyone ever compiled a Linux version from your source code? I'm currently considering setting up a Linux box for audio encoding, storage and serving purposes.

Actually, would you know anything about C programming, by any chance?

David is looking for someone to port WavPack to Linux. Should be really easy, because the code is 100% ANSI C. Still, some coding will be needed.

Hopefully, some linux code will be available soon.

Regards;

Roberto.

I do know a few things about C programming actually, but I haven't touched it recently.

I may look into this some more in the future, if someone else doesn't do it in the mean time, but I just don't have enough time at the moment to have a decent look at it within the next couple of months.

Den.

Den, I suggest that if you do further testing, you could make the tests blind more easily.

Once you have the wavs decoded, randomize them blindly, by someone else disordering them and changing their names, but taking note of which one was which. Then convert them to MD format, and listen then as usually. Write down your results, and then check which file was which.

Thanks KikeG.

Apart from the tests I've posted, I've since tried a few more combinations, and blinded these using a method not unlike what you describe. The randomise function in Foobar does a good job of this...I ended up placing the A B tracks at the beginning of the MD, and then randomised the rest (or rather my wife did) and it worked quite well. That way I could also go back and check the A and B selections myself, then try each X track. A reasonable approximation of trusty WinABX on my MD.

I can easily separate the original wav/wavpack lossy everytime from those transcoded from ogg/mp3 and those from mpc. Each type introduces a unique set of artifacts, as briefly discussed above. I can't always separate the vorbis from the mp3, or the wavpack lossy from the original wav, or the mpc q5 from the mpc q10, but I can always separate the vorbis/mp3 from the mpc from the wav/wavpack, whether I can see the songs being played, or not. Did a few tests with my wife's assistance, and got 15/15 everytime in terms of the above groupings.

Den.

Just Failed to ABX WavPack @ 320 with flying colours. 

21:21:08    0/1  p=100.0%
21:21:20    1/2  p=75.0%
21:22:30    1/3  p=87.5%
21:23:45    2/4  p=68.8%
21:24:14    2/5  p=81.2%
21:25:09    2/6  p=89.1%
21:26:10    3/7  p=77.3%
21:27:12    3/8  p=85.5%
21:28:12    3/9  p=91.0%
21:29:13   3/10  p=94.5%
21:30:14   4/11  p=88.7%
21:31:15   4/12  p=92.7%
21:32:15   4/13  p=95.4%
21:33:16   5/14  p=91.0%
21:34:17   6/15  p=84.9%
21:35:17   7/16  p=77.3%
21:35:29  test finished

[span style='font-size:8pt;line-height:100%'](More scary that I took almost exactly a minute to do each one)[/span]

I had absolutely no idea what I was looking for whatsoever.  And this particular track, Crash Test Dummies / Afternoons and Coffeespoons, I can ABX easy with 320kbps Lame.

Couple of other points to note:
::Lame (with Razorlame) took 3:10 to encode @ api, Wavpack took 0:43.  Musepac takes 1:44
::160kbps with Lame is 4,615KB vs 7,657KB with WavPack     
::320kbps with Lame is 9,229KB vs 9,230KB with WavPack
::Being as I cant ABX Musepac at all, no test was done for MPC vs WV     

Will try problem samples sometime soon...

I had absolutely no idea what I was looking for whatsoever. And this particular track, Crash Test Dummies / Afternoons and Coffeespoons, I can ABX easy with 320kbps Lame.

Couple of other points to note:
::Lame (with Razorlame) took 3:10 to encode @ api, Wavpack took 0:43. Musepac takes 1:44
::160kbps with Lame is 4,615KB vs 7,657KB with WavPack 
::320kbps with Lame is 9,229KB vs 9,230KB with WavPack

I can ABX wavpack @ 320 kbit everytime with all the music I have tested, 15/15. What I hearing is a very, very subtle increase in background noise (hiss). It is more obvious with music that has some "space" in it, such as a basic percussive introduction, so you can hear silence between the notes. To hear it, I need to have the volume up very high, and be in a very quiet room  (apart from my noisy PC) with decent head phones. I'm not hearing any other artifacts though, and I imagine that with certain audio hardware set ups, the hiss might be undetectable. At 352 kbits, or whatever budgie advocates, I can not always hear it, depending on the music I test with.

The reason for the different file size at 160 kbit, is that while Wavpack lossy will accept bit rate settings as low as 32 IIRC, the lowest number it can actually achieve is ~260 kbit. One track I tested got down to 230 kbit. If you set numbers below this, it will just go as low as it can.

Den.

I couldn't resist trying wavpack with some of these infamous samples everyone keeps talking about, so I downloaded castanets.wav and fatboy.wav, and thought I'd give them a try. First I encoded them both with LAME 3.90.2 alt preset insane, and Wavpack lossy @ 320 kbits with the high quality switch.

Before proceeding, I checked them out individually, and I can see why they are used to stress encoders.

I started ABXing fatboy.wav, and never got to ABXing castanets, because the difference with fatboy caught me by surprise.

I immediately hit 15/15 with the original fatboy.wav vs LAME 3.90.2 insane. LAME has a good crack at it, but I can clearly hear some extra "warbles" in the middle of the sample, that stand out every single time.

Then I tried fatboy.wav vs Wavpack lossy @ 320 kbits, at the same volume as the above test, which was a comfortable listening volume. Hit 8/15, with WinABX. 

I then increased the volume right up, and focused on the beginning of the sample. Hit 15/15 this time. Again, a very slight, but perceptible increase in background hiss gave the wavpack 320 kbit sample away. The fatboy sound itself sounds essentially identical with my ears.

Finally, I created a new wavpack lossy file @ 448 kbits, aka as wavpack --preset budgie    and I found the very slight hiss was reduced, and I only got 10/15 with Win ABX. Tried again, and got into the "zone" and hit 14/15.

Please note that these were encodes, not transcodes like my previous posts. I'm thinking that if someone is considering LAME insane as an option, but they don't need mp3 compatibility they should also consider wavpack lossy @ 320 kbit, because it may handle the problem samples better, although this needs more testing. The barely perceptible hiss is the only thing that is allowing me to ABX Wavpack at these bitrates, and I can only sense it at high listening volumes.

Den.

Tried with castanet2.wav file :

255 kbps (mesured), high quality

---------------------------------------
1L File: G:\Temp\casta2\castanets2.wv.wav
1L Rating: 3.4
1L Comment: 
---------------------------------------
ABX Results:
Original vs G:\Temp\casta2\castanets2.wv.wav
    14 out of 16, pval = 0.002

321 kbps (mesured), high quality - focused on beginning of the track

---------------------------------------
1R File: G:\Temp\casta2\castanets2_355.wv.wav
1R Rating: 4.2
1R Comment: 
---------------------------------------
ABX Results:
Original vs G:\Temp\casta2\castanets2_355.wv.wav
    8 out of 16, pval = 0.598

THE SAME, 321 kbps (mesured), high quality - focused on the middle of the sample [intermediate attacks]

---------------------------------------
1R File: G:\Temp\casta2\castanets2_355.wv.wav
1R Rating: 4.2
1R Comment: 
---------------------------------------
ABX Results:
Original vs G:\Temp\casta2\castanets2_355.wv.wav
    21 out of 30, pval = 0.021

(10/16 on the middle of the test, but I put up to 30, because of a really bad beginning)


First test was easy : pre-echo on the first and strong attack.
Second was more problematic : I'm not sure of what I heard. Sometimes pre-echo, sometimes different tonality (or is it added noise ?).
In the two cases, result is really good. Artifacts with wavepack lossy seems to be in general completely different from others perceptual codecs : noise, and on very rare cases, pre-echo. I can't perceive anything else wrong. At least, for the moment...

Thanks for your post guruboolez.

It's good to see some other tests come through from yourself and Lev help reinforce my own findings so far about the potential for Wavpack lossy. It certainly behaves differently from the other codecs.

Den.

Thanks for the results; they're very interesting.

Am I correct in my understanding that a large part of WavPack's compression is basically reducing dynamic level quantization (that's what I gathered from the comparisons to how lossless codecs compress quiet sounds well)?  If so, that'd explain the slight hiss, as what it's basically doing is raising the noise floor (or lowering the S/N ratio, if you prefer to think of it that way).

Theoretically this should be worse than psychoacoustic codecs, since it's dumping all the bitrate savings into degrading the S/N ratio, while psychoacoustic codecs should get some "free" bitrate savings by dumping inaudibly masked sounds.  But if the psychoacoustic model has significant flaws (as nearly all do, in one way or another), WavPack's method might actually get you more consistent results; i.e. it won't be as good on samples where the psychoacoustic codec does well, but it won't ever fail as badly either.  This would also explain the better transcoding performance, as most bad transcoding artifacts are psychoacoustic artifacts, not S/N ratio problems.

I think it's really good that I could contribute a little to awake an interest in WavPack HQ lossy mode. WavPack is great tool to archive music and really deserves a bit more care than it has had so far. Especially those, who play with thoughts for later transcoding, should focus on WavPack HQ lossy mode... I think later they will appreciate it.

Den seems to be just amongst the few here at HA, who use their own brain and don't just repeat everything like parrots (like transcoding is evil and another stupidities)... And besides, he loves MiniDisc, just as I do, although I know it's not the best quality available; but it's small and convenient... and its quality is not such horrible as some here would say, especially when you not compare it with original source of sound. 

For my part it's needed to say I did a lot of exhaustive and thorough test with WavPack and many other formats as for transcoding, but I somehow feel my presence and my opinions are not welcome (just wonder why...  ) I did a lot of ABXing, but as I am not trustworthy person for some people (just wonder why, again...  ), I didn't bother with posting the results. Although I bet with everyone my hearing is excellent and although not trained for hearing artifacts, it's trained damn good to listen to subtleties and spot even the delicate differences between original and copy.

Sorry, den, you could spare a lot of work or you could make maybe something more useful... but at least you made your own experience and know far better now, I hope so.

That's approx. all... for now 

This is a very interesting thread. One interesting thing: when you've been "in to" psychoacoustic coding for a long time, the slightest audible artefacts become really annoying - whereas plain uncorrelated background noise or hiss can be quite acceptable. Of course, I'd rather have perfection, but given the choice between artefacts (which are usually correlated with the signal) or background hiss, I'd rather have hiss any time!


This reminded me of a comparison I made a few years ago: I compared wavepack lossy (but not HQ - I didn't know about this switch) with the "very BETA I'm not supporting this you can't get it anymore" lossy monkey's audio. I found lossy monkey to be better i.e. I could hear the hiss in wavepack lossy, but not in monkey's audio lossy. Unfortunately, I can't remember what bitrate I used for wavepack lossy, but it was probably 256kbps. As such, this comparison was probably useless, but I think I'll go back and try it again - if I can find lossy monkey.

Cheers,
David.

P.S. lossy monkey was an idea that the monkey's audio author had, but quickly rejected. It produced VBR files which held all the information used to reproduce the "predictable" parts of the original file, but not any of the residual "noise-like" parts which had to be stored "as is" to make conventional monkey's audio files lossless.

Just for the record, shorten has a lossy mode too.

Here is that tread again David, with posts about lossy ape and wavpack almost 2 years ago:
http://www.audiofora.com/yabbse/index.php?...d=1687;start=15 (http://www.audiofora.com/yabbse/index.php?board=4;action=display;threadid=1687;start=15)
Fun to read again!

I can put that 'lossy ape' package online in a few hours, for those who like to experiment with it. It has no level switches and gives bitrates averaging a bit below 400 kbps IIRC.

I'm thinking that if someone is considering LAME insane as an option, but they don't need mp3 compatibility they should also consider wavpack lossy @ 320 kbit, because it may handle the problem samples better, although this needs more testing.

One thing to take into account are that the problem samples are specifically geared towards classical lossy codecs. They don't necessarily are problem cases for WavPack (more or less confirmed by your testing), which might have another class of problem samples.

From what I've seen so far, I'd think classical music with wide dynamics might be a problem for WavPack, compared to classical codecs. Maybe I'll do some tests

, but I somehow feel my presence and my opinions are not welcome (just wonder why... 

...because you make wild statements you do not back up with any data or testing.

den did the opposite, he posted his testing data and allowed us to make a conclusion, offering his own in the passing.

It's a major difference, and the reason why I take him seriously and you not.

Here is that tread again David, with posts about lossy ape and wavpack almost 2 years ago:
http://www.audiofora.com/yabbse/index.php?...d=1687;start=15 (http://www.audiofora.com/yabbse/index.php?board=4;action=display;threadid=1687;start=15)
Fun to read again!

Yes - it's amazing how useful some of the information is at the old r3mix forum - and it's not even properly linked from r3mix.net anymore. I hope it's archived properly somewhere. Thanks for finding that!

I can put that 'lossy ape' package online in a few hours, for those who like to experiment with it. It has no level switches and gives bitrates averaging a bit below 400 kbps IIRC.

It would be great if you could - I can't find my copy and I've searched several CD-Rs which should contain it. I don't know what Matt would have to say though.

The HA community would be a great place to figure out if it really is as transparent as I thought at the time - I didn't have a quiet listening environment, so I would have missed low-level hiss. But it had measurably less hiss than Wavpack at the time, so maybe...

Cheers,
David.

It's a major difference, and the reason why I take him seriously and you not.

You're young, Garf (AFAIK), and so I forgive you... When you get into post-mid-age and be a bit more experienced man, then you will understand a lot more of things you don't understand today. Maybe... 

Hmm, it seems the lossy ape dll (from may 2001, version 3.89) can't work any more under monkey's audio 3.97. But fortunately MAC 3.92b2 is still online: http://www.monkeysaudio.com/files/MAC_392b2_F.exe (http://www.monkeysaudio.com/files/MAC_392b2_F.exe)

Well, the fact that you'll need to uninstall, install again etc. will ensure only the very motivated here want to go through this! So here is the lossy macdll.dll and winamp plugin: http://web.inter.nl.net/users/hvdh/MAC_Test_01.zip (http://web.inter.nl.net/users/hvdh/MAC_Test_01.zip)
Check the readme.txt first. At startup you'll read "Invalid DLL Version" but then just click OK and continue.
Remember it's slow and forever unsupported. Please don't bug Matt about this! Personally, I'm just curious if wavpack has actually improved in the meantime.

Hans

For my part it's needed to say I did a lot of exhaustive and thorough test with WavPack and many other formats as for transcoding, but I somehow feel my presence and my opinions are not welcome (just wonder why...   ) I did a lot of ABXing, but as I am not trustworthy person for some people (just wonder why, again...   ), I didn't bother with posting the results.

The rules are the same for everybody. Den did blind testing and presented results. This doesn't work so that we start believing some people without any blind testing results and some other people only with blind testing results. Everybody is in equal position in that sense, and it's also the only fair way to handle testing.

The critique which you may have encountered is exactly because you have not presented blind testing results. That's pretty pathetic excuse that suddenly you did lots of ABX, but you "didn't bother posting", although all the critique has been due to lack of any clearly documented blind testing results by you. 

First of all, I'd like to thank everyone who has taken the time to test WavPack and report the results here. And, of course, thanks to Budgie for his flagrant disregard for rule #8!

There have been several comments about the "torture" samples and what is being said is essentially correct. The samples that are very hard for the other codecs might be easy for WavPack, and the other way around. Unlike conventional codecs, WavPack lossy does well with very complex music with lots of frequencies because the added quantization noise is more easily masked. Also, it does better with transients because it cannot have any pre-echo or smearing in the time domain (I'm not sure what you were hearing Guruboolez) because there is no digital filtering.

However, it does not do as well with simple music with limited frequencies because the noise gets added in parts of the spectrum that might not have enough masking signal. I think this is what Den is talking about when he mentions hearing the noise in the "space". Advanced codecs have the advantage here because they can shuffle bits around to concentrate on specific parts of the spectrum. The worst case I have found is that in electronic music there is sometimes significant energy at very high frequencies that can can raise the noise floor significantly. This short sample from the Fast and the Furious soundtrack is an example:

www.wavpack.com/Furious.wv (http://www.wavpack.com/Furious.wv)

The second "bleep" in the sample has an enormous amount of complex high frequency information. The default WavPack 320 lossy mode fails miserably on this. Using "high" mode (-h) and joint stereo (-j1) help considerably, but it's still not transparent. I suspect that other codecs would probably do better because all the added noise would be up high, although I have not tried any of them.

2Bdecided:

David, I believe you are correct about the relatively benign effects of uncorrelated background noise. Unfortunately, the noise added by WavPack lossy is slightly more complicated because, while it is ruler flat in frequency, it is modulated in level by the audio. In some sense this is better because it can be much more easily masked in quiet parts of the music, but it has the disadvantage that its changing level can draw one's attention to it (like in the old dBX noise reduction system).

BTW, there have been significant improvements to WavPack's lossy mode since you tested it in comparison to lossy monkey, including the addition of the hybrid mode and more refined predictors. I just ran a quick comparison and the latest version achieves about 10 to 12 dB less noise (at the same bitrate) than the version from that era.

Garf:

I recall that RKAU has a lossy mode too, although I have never tested either entensively. WavPack has had some sort of lossy mode almost since the beginning, but it really wasn't until the hybrid mode was introduced (which was based on Matt's original idea) that there was any appreciable interest.

Thanks again, guys!

Also, it does better with transients because it cannot have any pre-echo or smearing in the time domain (I'm not sure what you were hearing Guruboolez) because there is no digital filtering.

Are you sure ? Try with the exceptionnal and unatural killer sample called short_block_test2.wav
I tried to ABX it with wavepak lossy HQ@250 kbps (mesured at 178 kbps), and succeed. Didn't try with lossy HQ 400 (198 kbps mesured), but did some capture with CoolEdit :

http://membres.lycos.fr/guruboolez/images/...ck/original.png (http://membres.lycos.fr/guruboolez/images/wavepack/original.png)
http://membres.lycos.fr/guruboolez/images/...wvlossy_250.png (http://membres.lycos.fr/guruboolez/images/wavepack/wvlossy_250.png)
http://membres.lycos.fr/guruboolez/images/...wvlossy_400.png (http://membres.lycos.fr/guruboolez/images/wavepack/wvlossy_400.png)

The time smearing I heard on ABX test appears on the second screenshot.

However, overall performances are extraordinary for this bitrate. Musepack --standard reach the 1000 kbps ! (but is transparent, at least for my ears). Ogg GT3b1 -q 5 is near 500 kbps... Don't try with --alt-preset standard, but I suppose it to be close to 320 kbps ceiling (and with heavy pre-echo).

Bryant, I made a lossy ape (http://web.inter.nl.net/users/hvdh/Furious.ape) of your sample, and made a wv with settings -b400 -h. Filesizes are nearly equal. Still I find the wv add hashness that the ape doesn't. Please try if you agree, you just need replace the in_ape of winamp temporarily with the one of my link above (eh, you DO have in_ape already, don't you?  ).

If you read the 2-year old post, you'll see already then I hoped Matt would pass his methods to you!

Have you also thought about filtering the audio with some 'inversed' ATH curve, to put more weight to extremely high- or low-frequency content before calculating the energy and trowing away bits accordingly? (I hope I make this thought clear!)

Hans

Edit: Removed stupid question related to enabled crossfeed plugin messing with me once again!

LOL.. now bryant is dragged to the lossy quality business.. poor guy.. 

Bryant, the -j1 switch you mention indeed gives much improvement at least on your 'Furious' sample, 'udial', and I suppose all (near) mono samples in general. Would you recommend to just always use it?

Quote

It's a major difference, and the reason why I take him seriously and you not.

You're young, Garf (AFAIK), and so I forgive you... When you get into post-mid-age and be a bit more experienced man, then you will understand a lot more of things you don't understand today. Maybe... 

And the fallacies abide....

It's rather ironic that you mention "thinking" in your earlier post.  The statements you continue to make surely do not seem to be a product of critical thinking.

Guruboolez:

I tried to find that sample but could only find one copy and it was compressed with something I couldn't figure out. Could you upload something for me (or point to where I might find it)?

I see what you are talking about, but it seems like it only occurs after the burst, right? This makes sense because there would be some trailing quantization noise there (which could sound like smearing), but there would not be anything before the burst (which is what I thought pre-echo was). I would still like to hear this sample, though.

HansHeijden:

Actually, Matt and I were in frequent contact during that time and traded ideas often. His lossy version was something that he put together in a couple hours IIRC and was really very simple. His lossless encoder was already dividing the audio into blocks, and so what he did was when he encoded each block he checked the level first and then made a decision as to how many LSBs to strip off all the raw samples before he compressed the whole block losslessly. The bits stripped off would obviously go into the "recovery" file (which I don't think he ever implemented). At around 400 kbps this was often transparent, and because his lossless compressor was so good it was very efficient.

The problem with this is that the quantization noise, while very low, can be jumping up and down 5 times per second in 6 dB steps. As long as it's completely inaudible that's fine, but as soon as it approaches the threshold of audibility you might start hearing those discontinuities and I suspect that they would be pretty obnoxious.

WavPack's lossy mode is implemented much later in the compression process and filters the quantization noise rate of change to prevent any audible jumps. I believe that this allows WavPack's lossy mode to be acceptable at lower bitrates because even when the noise approaches audibility it doesn't make itself too obvious. Matt and I discussed this and he decided that incorporating a similar scheme into Monkey's Audio would involve changing too much (and I think he didn't want to wander into the quagmire of lossy codecs).

There is room for improvement, however. For one thing, the lossy mode is currently strictly CBR. It would be possible to implement some simple psycho-acoustic algorithm to determine the relatve audibility of the quantization noise and adjust the bitrate accordingly (i.e. VBR).

Also, I have thought a little about some intelligent switching for joint stereo. As you point out, true stereo is the default and I did this because I found some samples where joint stereo generated some problems (loud sound in only one channel produces audible noise in the other quiet channel) even though most samples benefit from joint stereo. A simple algorithm could be devised to detect the problem cases and only switch to true stereo then.

JohnV:

To use a common American expression, I will try to avoid letting lossy compression become my Vietnam. 

Here is the sample, compressed by a simple but impressive ZIP algorithm :

http://membres.lycos.fr/guruboolez/AUDIO/s...lock_Test_2.ZIP (http://membres.lycos.fr/guruboolez/AUDIO/samples/Short_Block_Test_2.ZIP)

@HansHeijden
Thanks for making me aware of the lossy ape "beta", and the discussion around it a couple of years ago. I was very interested, and as I don't have Monkey's installed, I fired this one up and also checked out bryant's furious sample. I agree that it certainly gets the sample more "correct" than Wavpack. I could still ABX the Monkey's lossy version 10/10, but it was very close, and I was mainly sensing different levels of background noise, rather than artifacts in the "music" itself. I also tried Wavpack lossy at 400 kbit to more closely match bitrates, and it was much better than at 320, but not as close as Monkey's.

@bryant
Thanks for your comments about wavpack's performance and the sample itself. It was good to check out a sample that does trip it up so that I can be aware of potential shortcomings in Wavpack lossy.

Until now, I've been sticking to true stereo for lossy encodes when I am not keeping the correction file, because I was under the impression that the joint stereo may cause the odd problem. (I use Speek's front end with the high quality option checked, but as far as I know, it is not using the joint stereo switch.) If there was an easy way to make the joint stereo a safe joint stereo, without huge effort on your behalf, it sounds like a good idea. Otherwise, I might start trying some joint stereo encodes and see what happens. I've been very happy with the results so far though as it is. It's just that after checking out your furious sample, I'm starting to wonder about using joint stereo. Can you elaborate more on how risky it is? 

@2Bdecided
You hit the nail on the head with regards to why I think Wavpack lossy has some potential. The additional noise, or lower S/N is still very reasonable, and I would much prefer a slight increase in background noise if the recording itself is still of "high quality" whatever that means, and less or no artifacts. When it's music you don't know intimately, the odd artifact is no big deal, but when you know the music shouldn't sound like what you are hearing, it get starts to get in the way of enjoying the music. 

The huge benefit to me has been it's transcoding into ATRAC3 performance. Wavpack kills the other perceptual codecs consistently for this, and Delirium's explanation makes sense.

I'm going to check out this other sample of guruboolez out of interest now.

Den.

PS: @bryant, Sorry to drag you into the world of lossy, but I can't seeing this being your Vietnam somehow. Even if you don't develop this any further, I suspect you don't have much to worry about 

Just did a quick and dirty ABX test of wavpack lossy using the short block test 2 sample. Just used the high quality option, and didn't touch the joint stereo switch.

Went staight for Wavpack @ 400 kbits to see what it would be like. First tried to ABX the first third of the sample, hit 8/15. Then tried the middle section, and hit 12/15. Can't exactly describe what I was picking up, as they were very, very close, but the "tone" as it increases in pitch, did not quite sound/feel the same. I suspect that just as with the furious sample, wavpack lossy adds some extra high frequency artifacts, or something, and I was sensing this during the test. It was very hard to pick up though. 

I'd be interested to hear what you think about this bryant, as to whether it is a pre-echo/smearing effect as guruboolez suggested, or something else.

Den.

Guruboolez:

Thanks for the sample. I didn't realize that it was an impulse test before and those are a very interesting case because they are extremely simple in the time domain (which is why it zips so well) but very complex when converted into the frequency domain (which is why conventional lossy codecs choke on them). I was able to losslessly compress it with WavPack to about 200 kbps, so I suggest that you use that if you listen to a lot of these. 

When this are compressed with the WavPack lossy mode, the impulse is perfectly reproduced but then there is a little bit of residual noise that trails off as the predictor "recovers", and it's the low frequency portion (< 1000 Hz) of this noise that is audible, I believe. Turning on noise shaping (-s1) eliminates the low frequency portion of the noise and made this transparent for me, although like you said it was pretty close before.

I also tried lossy Monkey on this and it looks like it is simply stored losslessly, which is what WavPack should do too since there's enough bits to do it. However, I also noticed that lossy monkey introduced a DC offset, so I think there might be a bug in there somewhere.

Den:

I wouldn't characterize joint stereo in lossy WavPack as "risky", in fact it was the default in early versions because it generally results in lower noise and better lossless compression ratio. However, I did run into a few samples with stuff panned hard to the sides (i.e. early Beatles) in which I could hear cross-channel noise in headphones. I decided that I would rather sacrifice a little average performance for improved "worst case" performance, although now we are seeing a few "worse case" samples that favor joint stereo.

You are correct that Speek's frontend does not use joint stereo by default, but you can always add either -j0 or -j1 to force it either way. You might also want to experiment with noise shaping (-s1). I find that at low sampling rates (< 50 kHz) this generally causes the noise to be more "hissy" and audible, but it definitely increases the S/N ratio in the midrange. I would be interested in how you hear the difference.

Unfortunately, the current WavPack format supports only a single stereo mode for an entire file, so there would not be any way to add a "smart" mode at this time. However, WavPack 4.0 will allow on-the-fly switching, so it will definitely be possible then.

BTW, did you receive the e-mail I sent you last week through HA?

Thanks for the info bryant.

I concur that the joint stereo switch -j1 seems to benefit all of my everyday music samples at the moment. I was considering testing it using a early Beatles rip myself, because I have one of those early Beatles albums on CD with some instruments recorded only on the left or right channel. (Very early example of multitrack recording!)

Certainly with more modern stuff where there is a lot of hard stereo panning, eg New Order - Bizarre Love Triangle, -js1 still works well, I'm guessing because there is plenty of other audio still there to mask any potential increase in noise.

I'll check out the noise shaping switch as well but probably after the next two weeks. I'm about to head off on a holiday and do a heap of diving off Ningaloo reef north of here. I'm stoked to be escaping the stress of work emails, but also wondering about getting by for 2 weeks without keeping up with hydrogen audio. 

I just checked, and I did get your email, thanks for that. The email address I submitted to HA doubles as my spam email address for registering on web sites, so I don't check it as often as I should. I will do so more often now, because there were 1 or 2 other HA ones there as well!

I, and many others I'm sure are looking forward to Wavpack 4.0. Thanks again for this interesting codec.

Den.