Wavpack lossy

Topic: Wavpack lossy (Read 27471 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

Wavpack lossy

2006-09-19 16:50:54

Hi to everyone here!

As you can see i am a new one here. And i already need help. I would like to store my whole collection of cd´s to Wavpack lossy. Lossy because i am limited with space right now. My collection is jus to too big to fit (around 250 cd´s). And i don´t need prefect quality. For that i play cd´s on a hi-fi. But i need better quality that mp3 gives. So i am asking what is the best setting for lossy. EAC is ripper.

Thanks

Sorry for my bad english.

Wavpack lossy

Reply #1 – 2006-09-19 17:07:56

Why do you need "better quality than mp3" ?
Since you're concerned about space you should seriously try ABXing some LAME-generated MP3s at various quality levels and see whether you're able to hear any differences or not. I think you'll be surprised.

<addressing other forum users as well>
As for high bitrate lossy encodings: Is there really a demand for high bitrate lossy-only (no hybrid/dual-stream)? Who uses such encodings and why?
</addressing other forum users as well>

Wavpack lossy

Reply #2 – 2006-09-19 17:17:29

I think halb27 will provide solid arguments towards WV lossy, me... I can't ABX LAME -V0 --vbr new and WV -hx3b320m...

Wavpack lossy

Reply #3 – 2006-09-19 18:13:25

IMO using ~ 350 kbps with wavPack lossy yields an excellent quality which should be undistinguishable from the original for the vast majority of tracks. In the rare cases where differences are audible they should only affect the perception of noise (I know just 1 sample where this is the case (badvilbel): the existing noise in the original sounds very slightly different).

But you can use way less than 350 kbps and still get an excellent quality. However when going significantly lower I think you're better off using mp3 at ~250 kbps which too yields an excellent quality in a robust way.

It is advisable to use the -h switch (high quality mode) as well as the -x switch (uses more effort on the encoding side). For the best you can do you use -x6 instead of plain -x but be prepared thet encoding time is veeery long.

You can change the default noise shaping behavior. Any specific noise shaping has the problem that it can be good for some trracks and bad for others. Which makes it difficult to get at a solid experience on noise shaping behavior. According to shadowking's experiences the default works best with relatively low bitrates. At a higher bitrate like 350 kbps quality is consistently good when not using noise shaping at all (-s0 switch). My own experience confirms this (but is based on rather few tracks).

David Bryant is about to provide us with a new version of wavPack (alphas are available already).
This will give us a new -h mode which will be less CPU demanding than the current -h mode. This is important for mobile DAPs. The -x switches will change too.
This is very promising and the results I heard from the alphas were very good.

As for my high quality wavPack lossy and mp3 settings see my signature.

Wavpack lossy

Reply #4 – 2006-09-19 20:06:17

Quote from: halb27 on 2006-09-19 18:13:25

IMO using ~ 350 kbps with wavPack lossy yields an excellent quality which should be undistinguishable from the original for the vast majority of tracks.

For most people, lame -V2 is indistinguishable from the original on the vast majority of tracks, so I really suggest you try ABXing with a few different songs and settings before you waste a huge ammount of space. It's really not uncommon to fall for the placebo effect and think, for example, that mp3 sounds bad. We have a kind of "Audiophiles Anonymous" thread on this forum where people tell their stories of how they thought they had golden ears...

Wavpack lossy

Reply #5 – 2006-09-20 03:39:45

One real advantage of WavPack lossy (hybrid) is an absent of psy-model for sound compression. It will give better result if you'll want to transcode your high bitrate lossy files in something more compact (e.g. for portable device). We had discussion about this.
But if you really need smallest file with best quality and maximum portability - use recommended LAME and -V switch in 3..0 range (and --vbr-new too).

Wavpack lossy

Reply #6 – 2006-09-20 08:00:23

Quote from: iGold on 2006-09-20 03:39:45

One real advantage of WavPack lossy (hybrid) is an absent of psy-model for sound compression. It will give better result if you'll want to transcode your high bitrate lossy files in something more compact (e.g. for portable device).

Absence of a psy-model is NO advantage -- not even for transcoding purposes!
I already questioned your "reasoning" in the other thread and gave a more likely explanation of why transcoding test results showed thah high bitrate WavPack lossy is superior when used as transcoding source. It's not clear why this should have anything to do with the having-psy-model property.

Wavpack lossy

Reply #7 – 2006-09-20 09:56:32

Quote from: MedO on 2006-09-19 20:06:17

For most people, lame -V2 is indistinguishable from the original on the vast majority of tracks, so I really suggest you try ABXing with a few different songs and settings before you waste a huge ammount of space. It's really not uncommon to fall for the placebo effect and think, for example, that mp3 sounds bad. ...

I am used to abxing, and from that -V4 usually is transparent to me.

My personal demand is for robust high quality even for difficult tracks, and as disc space isn't a big issue to me I can allow for pretty high bitrates.
In order to find out what encoder and setting is good enough for me I use the three samples harp40_1, herding_call, and trumpet. Other than many pre-echo samples they are all natural music of a pure tonal character, and when listening to the original it is hard to beleive they provide such big problems to many encoders (nearly universally true for harp40_1, not quite so for trumpet and herding_calls which however are not restricted to current Lame though Lame has a special problem with them which so far is partially solved with 3.97b3 and 3.98alpha).

For these samples -V2 is not sufficient at all to achieve a good quality. I guess nearly everybody can hear that even without abxing. For any mp3 encoder it takes a setting with an average bitrate way beyond 200 kbps. VBR is suspected to do a bad job on it (experienced with FhG encoder from MMJB 6 and Lame encoder versions 3.90, 3.91, 3.96, 3.97, 3.98 (3.98 however is about to improve things) - Helix with level's setting however is fine). And when going very high average bitrate there is no much sense using VBR anyway. To me abr ~ 250 kbps is the best way to go, but cbr 256 kbps is good as well. 224 kbps should be minimum IMO.
As for wavPack lossy it takes ~ 350 kbps to make me satisfied with these samples.

This is my very personal procedure. It provides me with an easy way to find out about encoder settings which give me a certain security against bad encodings even for difficult cases. I don't know a better way to do it.

Quote from: iGold on 2006-09-20 03:39:45

Absence of a psy-model is NO advantage -- not even for transcoding purposes!
I already questioned your "reasoning" in the other thread and gave a more likely explanation of why transcoding test results showed thah high bitrate WavPack lossy is superior when used as transcoding source. It's not clear why this should have anything to do with the having-psy-model property.

I agree that the mere absence of a psy-model is no advantage. But to me it is an advantage that there is no filter-bank and no converting into the frequency-domain with it's restriction to temporal resolution especially the way it is done with the mp3 format. I'm well aware that this too isn't advantage in itself, and my usual tryout samples have shown me that wavPack lossy below 300 kbps can be pretty bad - doesn't sound just like added noise but sounds real distorted. So wavPack lossy is adequate to me only when using higher bitrates like ~ 350 kbps.
As for the trancoding quality wavPack lossy is suspected to provide a good basis according to experiences of den AFAIK. There was a methodological good transcoding test done by guruboolez some time ago in which wavPack lossy didn't come out on top but it was done at ~ 250 kbps which IMO is too low for very good quality wavPack lossy.

EDITED: '... that wavPack lossy below 300 kbps can be pretty bad' read '... that wavPack lossy below 300 kbps is pretty bad' which was true for these samples but gave a wrong impression when reading.

Wavpack lossy

Reply #8 – 2006-09-20 14:34:11

Thanks hlab27. You helped me a lot. I will try your settings. In the future i will convert all to lossless for backup. I want lossy just for listening on computer and for doing some compilations for car. Lossy is good enough for listening in car burned as audio cd (compilations).
Thanks.

Wavpack lossy

Reply #9 – 2006-09-20 14:43:24

Quote from: SebastianG on 2006-09-19 17:07:56

<addressing other forum users as well>
As for high bitrate lossy encodings: Is there really a demand for high bitrate lossy-only (no hybrid/dual-stream)? Who uses such encodings and why?
</addressing other forum users as well>

I use wavpack around 320 k

- No pre echo or any psycho acoustic artifacts. Worse case slight noise.

- Still very acceptable bitrate for lossy

- Transcodes transparently to any lossy codec . abxing reference lossy vs transcode from wavpack is usually not possible even in quite room with Grado headphones.

- Quality is very stable - as good as when I was using mpc.

- multichannel capable for future use. Good seeking.

- Opensource.

- No genious required for psymodel tunings. Not much to tune. This is where some people think the whole thing might be some toy to play with and not ready for serious use, however there is a strength in its simplicity.

- No recommended versions. All encoders are nearly identical quality.

- no gapless issues, clipping, stereo image shifts and other weird stuff that can sometimes happen with transform coders - (search vorbis aotuv white noise appearing in a sample from guruboolez )

- A lossy codec can become obselete, but these are also lossless codecs. Lossless codecs are more future secure.

Wavpack lossy

Reply #10 – 2006-09-20 14:59:32

Quote from: ElephantTLK on 2006-09-20 14:34:11

Thanks hlab27. You helped me a lot. I will try your settings. In the future i will convert all to lossless for backup. I want lossy just for listening on computer and for doing some compilations for car. Lossy is good enough for listening in car burned as audio cd (compilations).
Thanks.

In that case the hybrid feature is handy. Encode to lossless and create lossy file at the same time. Also -hx6 is probably useless as many sample don't gain extra quality from this - in fact sometimes it results in worse compression. Bryant has already mentioned this. Only limited samples benefit and its ssslllloowww. I think the wavpack website is a pretty good guide.

You might want to do some listening tests with mp3, vorbis, aac at common bitrates as they may be already sufficient for you at lower bitrates and are supported by hardware players.

Wavpack lossy

Reply #11 – 2006-09-20 15:24:03

Quote from: shadowking on 2006-09-20 14:59:32

... -hx6 is probably useless ...

Yeah, -hx6 may be way too much. I just wanted to get the best out of wavPack. I was always content with -hx6 but with pure -hx this probably would have been the same.

With the next version there is even less to decide with the new -x modes.

Wavpack lossy

Reply #12 – 2006-09-20 16:55:26

Quote from: halb27 on 2006-09-20 09:56:32

I agree that the mere absence of a psy-model is no advantage. But to me it is an advantage that there is no filter-bank and no converting into the frequency-domain with it's restriction to temporal resolution especially the way it is done with the mp3 format. I'm well aware that this too isn't advantage in itself, and my usual tryout samples have shown me that wavPack lossy below 300 kbps is pretty bad - doesn't sound just like added noise but sounds real distorted. So wavPack lossy is adequate to me only when using higher bitrates like ~ 350 kbps.

Those samples don't tell you much - they are good at delusions. A weaker encoder like wma or atrac *might* perform strongly on your test set samples, yet fail on many others. Maybe even vorbis or aac or even a really old mp3 encoder.

I don't think wavpack 250 k is weak at all when considering a wide range of samples. Its as good as at least 130k vbr on the normal coders. Various samples could be abxed on each encoder. Wavpack will win on pre echo etc and in other cases be worse on hiss. Worse cases might not sound great with any encoder at these 'low' bitrates. With smart noiseshaping its even more competitive.

I still prefer wavpack at this sub-optimal bitrate than what the other encoders do at their respective lower bitrates.

Wavpack lossy

Reply #13 – 2006-09-20 21:52:40

Quote from: shadowking on 2006-09-20 16:55:26

Those samples don't tell you much - they are good at delusions. A weaker encoder like wma or atrac *might* perform strongly on your test set samples, yet fail on many others. Maybe even vorbis or aac or even a really old mp3 encoder.

To be precise I should have stated explicitly that I would only consider a quality setting that is widely considered safe for the vast majority of tracks. For the security margin I use these samples.
BTW I've tried quite a lot of encoders on these samples and I always found to get at a satisfying quality it takes a considerably higher quality setting than usually is the case. Take for instance Vorbis (aoTuV 4.51). Usually -q3 is transparent to me (~115 kbps on average), but it takes -q5 (~160 kbps on average) to get at satisfying results with these samples for me.
Sure the procedure is oversimplifying things but it's a simple method of getting a security margin within the set of settings usually considered safe. Of course it is personal preference whether to take exactly that setting that gets excellent results for these samples or to provide for an additional safety margin or lower the setting a bit quality-wisee cause these samples are really extreme. And of course this specific set of samples is questionable. For more general use it would be wise to add a typical pre-echo sample like castenets - to me personally however that's useless cause I am pretty deaf towards pre-echo problems.

Take the procedure as a rule of thumb for finding a setting yielding robust high quality.

Wavpack lossy

Reply #14 – 2006-09-20 22:13:02

Quote from: shadowking on 2006-09-20 16:55:26

... I don't think wavpack 250 k is weak at all ...

I don't think so either but with a bitrate like that I'd prefer good old mp3 (or - if it's only about quality - vorbis or aac or mpc).

Wavpack lossy

Reply #15 – 2006-09-20 22:32:13

Pre-echo is specific to transform encoders; Wavpack lossy, ADPCM, DualStream are not affected by this problem.

It's very hard to tell which encoder family perform the best at very/ultra-high bitrate (> 300 kbps). Both have specific (and maybe opposite) issues: smearing for transform encoders; noise for the second type of encoders. WavPack/DualStream are handling without problems all samples on which AAC, aoTuV, MPC... are failing, and the latters are sometimes more transparent at 120 kbps on samples ABXable with Wv/DS at 350 kbps.

The result of any test will therefore entirely depend on the inclusion or exclusion of very rare problem cases. 10 "preecho-like" samples vs 0 "noisy" samples will show WavPack/DualStream as absolute winner; 2 "preecho" vs 8 "noisy" samples will prove that transform encoders are really better; and a "5 vs 5" fight will end on... no conclusion at all.

On one side I personaly had less troubles to collect ABXable samples for WavPack/DualStream at > 300 kbps (below, there's no debate in my opinion: transform coders are much cleaner): all consists on a subtle addition of noise. On the other side, artefacts with the other kind of encoders are sometimes more weird: I remember Serge Smirnoff iTunes AAC@320 kbps issue (minor, but as audible as unexpected), some strange noise packet suddenly appearing and brutally disapearing with Vorbis (creaking or Bloch), and also audible ringing with MPC -q10...

Wavpack lossy

Reply #16 – 2006-09-20 23:03:47

Quote from: guruboolez on 2006-09-20 22:32:13

... On one side I personaly had less troubles to collect ABXable samples for WavPack/DualStream at > 300 kbps ...

Can you give an example? I only have experience with badvilbel: more or at least an other kind of noise distinguishable from the original noise up tp 500 kbps (but I can easily make peace with it at 350 kbps).
I'd like to get more experience with samples where wavPack lossy behaves a bit poor.

Wavpack lossy

Reply #17 – 2006-09-20 23:45:00

My main computer is currently in move boxes - and can't access to my sample library. I'm not even sure that I kept them. Involved samples are always long tonal notes mainly coming from: wind instruments (flute, recorder, oboe, organ, accordion) but also some lyrical pieces and other ones (like a martenot waves sample).

The first problem samples I discovered were posted three years ago but aren't available anymore. Maybe some people have kept them. The thread is about the experiental WavPack lossy VBR but IIRC the released encoder (ABR) had similar issues with the same samples (especially the organ one which was really critical at this time).

Wavpack lossy

Reply #18 – 2006-09-21 08:03:51

Quote from: guruboolez on 2006-09-20 23:45:00

My main computer is currently in move boxes - and can't access to my sample library. I'm not even sure that I kept them. ....

Thank you very much. Don't search for them - just if you should happen to run upon them one day it would be nice if you could post them. As for samples with long tonal notes coming from wind instruments I'll look through my own musical collection a bit. (herding_calls btw has a long tonal note - and female voice is some kind of wind instrument , and I know it's an issue to wavPack lossy at a lower bitrate).

Wavpack lossy

Reply #19 – 2006-09-21 20:32:30

I still have those files for testing and have uploaded them here:

http://www.rarewares.org/wavpack/guru/

The idea that transform codecs do better with highly tonal samples than WavPack/OptimFROG lossy makes sense to me. The quantization noise added by WavPack lossy is basically flat across the spectrum (although it can have a general tilt up or down via first-order noise shaping). With highly tonal samples there's no place to hide the quantization noise, so you hear it above and/or below the desired audio. This is not good because the noise is audible, but because it's not all concentrated around the same frequencies as the music, it is easier to ignore. When transform codecs fail, you're going to find the problems right on top of the music.

On the other hand, wide-range material with energy all across the spectrum is relatively easy for WavPack lossy because the noise can be well below the signal at every frequency. In fact, WavPack lossy might be actually better with these samples than transform codecs (at high bitrates) because the total noise is less and there are fewer other things in the signal path to go wrong.

Wavpack lossy

Reply #20 – 2006-09-21 21:18:30

Quote from: bryant on 2006-09-21 20:32:30

I still have those files for testing and have uploaded them here ...

Thanks a lot, David. I will try them.

Wavpack lossy

Reply #21 – 2006-09-21 22:37:54

Just tried guruboolez' samples, and also found them pretty bad at < ~300 kbps, especially bruhns.
But at ~ 350 kbps (tried 4.31 -hx6, 4.31 -hx, and 4.44a2 -hx3) results to me are what I call not-at-all-annoying, and I don't think I will hear a problem in normal listening situations.

This is a good discussion as it points out with what kind of music the different encoding techniques have advanteges and disadvantages. And as the advantages and disadvantages are subjective, everybody has a basis for picking out 'his' encoder by trying the specific samples.
Thank you, guruboolez, for providing the samples and pointing out essential practical differences with different encoding techniques.

Wavpack lossy

Reply #22 – 2006-09-22 06:53:08

Quote from: halb27 on 2006-09-21 22:37:54

Just tried guruboolez' samples, and also found them pretty bad at < ~300 kbps, especially bruhns.
But at ~ 350 kbps (tried 4.31 -hx6, 4.31 -hx, and 4.44a2 -hx3) results to me are what I call not-at-all-annoying, and I don't think I will hear a problem in normal listening situations.

Same opinion for me. I said in my previous post that the artefact consists in a subtle addition of noise - not even easy to hear IIRC on ABX tests. What could eventually bother people is that artefacts at such high bitrate are almost impossible to catch with the all modern encoders (like aoTuV) or at least that problem samples are much more harder to find. Finding ABXable samples at 350 kbps was easier (at least with classical music) for WavPack lossy than for all other encoders.
Will the next improvements (VBR mode, different noise shaping) improve WavPack's transparency?

Wavpack lossy

Reply #23 – 2006-09-22 07:33:43

Tried bruhns 1st thing in the morning. Its really low volume - how to I normally listen to this ? replaygain or high volume ?

Encoded 256k -hx1. Didn't sound bad. Pumped the volume way up and found the spot around 8.6 secs. ok hit 8/8 - blanket hiss

320k with replaygain: 7/8 .. Came back home and went again for 320k and got 2/8.. went back to 256k and got 8/8 easily.

256 is obvious , 320 less so and requires a 'warm up' at 256 or really high volume. I don;t think its a problem unless you abx 256 first.

-x3 is not making much difference over -x1 like many samples I've tested. Is it really worth the extra encode time ?

noise shaping (negative) should help these samples. This hiss sounds different with -s-0.5, but still easily abxed - less hiss more 'blocky' and spread out. Haven't tried the alpha 4.xx encoder yet.

I have 2 samples similar to bruhns this in my collection that I can post

Wavpack lossy

Reply #24 – 2006-09-22 08:22:24

Quote from: shadowking on 2006-09-22 07:33:43

... I have 2 samples similar to bruhns this in my collection that I can post

Yes, I'd like to listen to your samples though I don't expect to get a new aspect. Anyway it's good to have more experience.

Taking it all together everything looks like for achieving a very good robust quality it is advisable to use 300+ kbps with wavPack lossy. And in that range it's as always with lossy encoders: given a certain setting in the robust quality range you can achieve an even more improved robustness by going higher with quality settings but in a practical sense the ratio robustness improvent vs. additional effort (file size, encoding time, ...) is quickly getting worse the higher the quality setting is already. It's just personal preference where to put the optimum point. It also depends on the application context. For me who doesn't listen to CDs any more and doesn't want an additional lossless archive (or additional correction files) I feel more comfortable using 350 kbps and additional to -h a high -x mode. (And when the first 16GB+ high quality flash players will come up I guess I will even go a bit higher - I just realized the iPod people have arrived at 8 GB for their nano). But this is just my personal paranoia. Your 320 kbps setting is certainly fine, and if it is only about playing music additionally to keeping a lossless archive or keeping CDs in an active state maybe I'd do that too - or be totally satisfied with very high bitrate mp3 as mp3 has additional desirable features.

Notice