For the future I plan to do my encodings using very high bitrate mp3 as a rule and wavPack lossy for very high quality cds.
I want to use it on my Rockbox powered DAP, so battery life is an issue apart from quality. Disc space isn't cause after having most of my music encoded with high bitrate mp3 just 6GB of my 40GB iRiver H140 are filled.
As for wavPack lossy I have some experience, but not very much, with problem samples as well as real life encodings from my musical archive.
As for that I know I do not want to go below 350 kbps, and I do not want to use fast mode. I also know badvilbel is a sample that requires 512 kbps to get transparent (to me).
On the other hand I know -b352x is sufficient to make any other problem sample I tried at least not-at-all-annoying, and usually this yields pefect quality to me with 'normal' music.
So -b352x.... is my favoured candidate so far.
I'm a bit uncertain as for noise shaping.
My own little experiments have shown that when quality is very high though not perfect, -s0.5 is slightly superior to -s0, and -s0 is slightly superior to the default setting. -s0 anyway is attractive because of it's being non-biased.
@shadowking (in case you read this - hope you do): you would prefer -b352xs0 over -b352xs0.5, wouldn't you? (and you have certainly more experience with wavPack than I have). And you would prefer l/r stereo over m/s stereo? Guess your advice is to use -b352xis0j0, right? Can you tell about your preference for l/r stereo?
For a precise question to everybody who'd like to help with his experience:
Can I do better than using -b352xis0 for the framework I described?
The default NS setting for 44.1khz is s-0.5 according to Bryant. In my experience it can make the hiss less 'sharp' on some samples, but make others more 'noisey' - sometimes dramaticaly. My feeling is that in a situation where it doesn't work , you might abx 350k samples that otherwise would be transparent.
To my ears the quality using no shaping and joint stereo off:
200k - abxable all the time
250k - abxable some of the time
300k - becoming transparent on most samples
350k - transparent > 99% of material
From this guide and your needs I'd go with -xs0b352
You get safety of bitrate + smart j/s. Another interesting bit is that -x6 can be good in some cases (herding, badvilvbel). Try -hb450x or -hb450x6 with vilbel - differences gone or are next to nothing. -h also helps on this 'sample'. Also on herding, harp etc , -x6 is slightly better.. One could use it with normal modes, but with high mode its painfull. Somehow I don't think its worth it for you as your setting is already pretty good.
Regarding joint stereo: You settings give smart j/s (-x) and can increase quality a bit (a lot on some samples). Without -x , the normal is -j1 for every frame, but on some samples with little stereo info -j1 can cause a noticeable noise increase. The other reasons I used -j0 was that I sometimes too lazy to wait for -x modes and I could test the 'base' quality of the encoder without enhancements. For normal listening and efficiency I think smart mid/side is the way.
The other questions in my head is DPL issues (I can't test that yet, but suspect no issues because there is no digital filtering in wavpack) and processing / transcoding. I got slight differences on 1 or 2 samples in transcoding abx tests. bitrate increase didn't help much, but -j0 made it perfect. I can't conclude anything because more i need more test cases, but still....I wonder.
Thanks for your hints, especially the x6 hint.
I tried -hx6 on badvilbel, and it really helped a lot. I've thought it over and over again, but in the end I will not use -hx6, as encodings are really sloooow, and I'm afraid -h has a serious impact on battery life of my DAP.
Instead I plan to use -b416s0x6. -x6 is better than -x on badvilbel also in normal mode (I could abx the difference). Usually x6 may be unnecessary, especially at 416 kbps, but as encoding speed is acceptable to me, there is no essential drawback.
-b416s0x6 yielded a good though not perfect quality to me with badvilbel, and I guess I can consider this an extremely bad sample for wavPack lossy. Moreover if I want to step ahead qualitywise from 270 kbps mp3 for cds with very good quality it sounds more reasonable to use 416 kbps instead of 352, and with this quality I can even consider not to keep an additional lossless file (or the correction file).
From former tests I know battery life isn't seriously worse with -b416 compared to -b352. Disc space isn't an issue to me anyway.
Thanks again for your help.
I tried this new optimized MMX wavpack encoder and it yields bit-identical results, with a nice speedup in -x modes. Give it a go if you are using the plain encoder.
Tell me, what is battery life for normal mode and high mode ?
high mode basically uses 100% of the CPU all the time, so I would imagine that the battery life is a noticeable amount shorter.
Thanks for the MMX version hint, didn't consider this so far. As I want to use -x6 this is valueable information.
Battery life was noticeably shorter on a former test when using -h. I can't tell about the exact amount cause I didn't record it.
But as I'm about to do something like 'final' decisions (okay: never say never again ) I should do another test. I don't like it very much cause it's pain (repeat the test several times as battery life changes a bit [guess battery doesn't get loaded to the very same degree each time], and as battery life is something like 9h IIRC there are some practical problems catching the very end of battery life if I start the test at home. Moreover Rockbox isn't very stable when I let the battery run down totally).
Anyway I will do another test, and report on it.
But I think it will take some time to do the test. Guess I'll finish it next week.
I finished my battery life test for my iRiver H140 using -b352hx6s0 and a non-high mode alternative which is supposed to achieve about the same quality. I chose badvilbel for finding the alternative (it is the only track where I can hear differences against the original at these bitrates). I ended up with -b412x6s0 as the alternative.
As I did assume I often missed the exact end when battery went down. On the positive side with current Rockbox firmware my H140 never crashed because of low battery. HD usually starts to stutter when battery is very low before the H140 dies, but that's all.
Battery life varied pretty much with each trial. Taking it alltogether battery life varied from 8h50min to 9h50min with
-b352hx6s0 as well as with -b412x6s0. I figured out that when continuing battery loading by drawing the charger cable and refitting it, and do this 3 times all in all, waiting each time until the loading LED turns off, I arrived at a rather constant battery life of 9h50min (with both -b352hx6s0 and -b412x6s0).
I did not expect this result. Thank you, shadowking, for motivating me to do another test.
Because of this result I will use b352hx6s0, because this makes the HD spin up more seldom (HD spinup is audible when music is quiet). The only bad thing is very slow encoding (it took me 5 hours with the MMX version to encode the 29 tracks I used for the trials). But I've made my peace with it as I'm not encoding new tracks very often. In those cases where I will do mass encodings, I can do it overnight.
Thanks again for all your help, shadowking.
halb27, shadowking, may I ask with which equipment you do the ABX testing to examine your settings:
audio amplifier and speakers or headphones, external DAC, directly on a DAP,...
I'll explain in length why i ask:
I found this knowledgeable forum through google and inspired by this particular thread, started my own testing. I like the idea of wavpack hybrid mode and use lossless on my audio server and lossy on my iriver H140 (which i do not own yet). Meanwhile, i've been testing with Shure E3 headphones directly on line out of my PC's motherboard (ASUS P4C800 deluxe).
I could ABX 16/16 -hb352xs0, by replay of 1 second pieces of badvilbel. Even harder but still ABX 16/16 with -hb408xs0. Keeping in mind the harddisk spinup remark, i'll stick with -b352. I also tested -x6, but to me it's not really an option, beeing 5 times slower than plain -x, even with the MMX wavpack encoder.
Now for my real concern: the basic noise on my motherboard is noticeable, and i'm afraid that the *very* acceptable -hb352xs0 results will disappoint me once i put the lossy files on my iriver H140.
All advice and insight would be welcomed.
edited: added -h to settings
I use Allesandro MS1 phones with TB santa cruz. Forget badvilbel. It doesn't even play properly in lossless. In Winamp it sounds totaly different in waveout and directsound. Also the -h or -hx improves it.
I suggest to test 6~12 normal samples from your own collection, the 128k test, even classical samples from guruboolez. Start at 300k and I think you'll already have a lot of difficulty abxing 8/8. On loud compressed stuff like metal, rock, punk - even 250k is often transparent or very close. I am talking about quite night abxing, so for portable use you have nothing to worry about.
Thanks for elaborating. I should have added -h to all above, which was what i actually tested, stupid me.
Anyway, it's reassuring to know that for portable use i can probably go lower. I'll check it out with real word samples.
One more question: with 'general' MP3 listening (i know that's not a very qualified statement), i noticed that besides the typical artifacts something else was wrong with the compressed music, which is hard to ABX on small samples: it's as if rythm is lost, the 'life' that distinguishes good from bad performers gets lost. Can anything resembling this be expected with lossy wavpack, if you know what i mean in the first place.
I think I know what you mean, but without abxing we can't say for sure. Because the transfrom coders have pre-echo and a variety of artifacts that actually *alter* the music when audiable, its possible to sometimes notice something wrong as the ear is very sensitive the first time and less later. I have sometimes abxed these suspicions. With wavpack its not to be expected because only hissing can happen and that can somehow 'blend in' and not change the music even in critical cases. You can wind the bitrate down to 200k & get a 'taste' then up it to 250k and the hiss will already start to blend in.
Well, i've been testing whole afternoon.
Seems that for all music i own and tested, i cannot distinguish any lossy file at 320k. It was already hard to find real world samples to distinguish at 200k.
Further more , one of the more time critcal music I know, Buena Vista Social Club, does not suffer at all at lower bitrates, so Wavpack seems indeed not to influence anything in the time domain.
The only reserve that I keep with respect to these findings, is the rather low fi audio output of the motherboard, an external high end DAC that i will use in the near future, will certainly reveal more, but for non portable use, i'll stick to lossless anyway.
Thanks for the help.
- With regular music wavPack 300 kbps quality is perfect to me too. I just decided not to keep a seperate lossless archive (or correction files) any more. This will simplify my life. And as for that I'd like to have a reasonable safety margin. That's why I will use 352 kbps.
- I use a Creative X-Fi internal soundcard. It has a superb quality to me and is more or less reasonably prized. As a headphone I use an Allesandro MS-2.
- If you can hear noise from your motherboard I don't think you have to fear your encodings are affected (but you can figure out by encoding to say hiqh quality mp3 and try this on any available DAP of decent quality. As for listening on PC a decent soundcard is supposed to help. I personally wouldn't consider to use a seperate DAC. I'd just find a soundcard that matches my needs.
- An iRiver H140 may be hard to find nowadays. Look for any DAP of very good quality that is supported by Rockbox firmware. The most attractive current DAP for this purpose may be the iRiver H340 as Rockbox has arrived at a rather mature stage for iRiver DAPs. An upcoming alternative is the Cowon iAudio X5 series - I don't know Rockbox's current state on it other than that it's usable already.
I'll have the exernal DAC anyway, as it's going to be a fun DIY project. And it will have a truly 'black' background, something i look forward too. When i finally am ready to test all this, i'll certainly report back here (might not be within the next year, you know how it goes with DIY)
Concerning the H140, i've recently ordered one on ebay and it's coming my way
It'll be interesting too, to compare the direct output of the H140 to the DAC (that's why i wanted the iRiver, for it's optical out)