Hi, I'm wondering what Ogg Vorbis "-q" setting people prefer for general computer playback. I'm not talking about really good audio equipment or archival purposes, in my case it's just for listening to (usually though headphones) from my computer.

Whatever works for you. I'd suggest doing a few encodes starting at about q5 and reducing quality until the artifacts become too annoying.

Yep I was expecting that response, fair enough though it is the logical solution.

I've just started playing around with ogg (I'm using the "aoTuV b3" Oggenc) and I'm amazed that it still sounds pretty reasonable all the way down to about q2.

I figure q4 is probably enough with my current equipment but I'm still interested in what others find acceptable or unacceptable for their purposes.

Personally, I find that q 4 is good enough for me. I dont have sophisticated listening equipment nor do I concentrate heavily on artifacts during casual listening. Also, the previous aoTuV beta shone at q 4.25 in the 128 kbps listening test, so it can't be too bad.

Personally, I prefer Q7. Q6 is sufficient for most songs, but I use Q7 as there are a few "problem" songs that still produce artifacts for me with Q6.

I tested the new aoTuV b3 with many samples of my music: jazz and rock in general... I don't hear differences with the original wave file in q6 setting in my ABX tests..., and not artifacts... but, I am a little paranoid with audio compression, then, I compress with q7

-q 4 is transparent for me through "headphones from my computer"

Normally I use 320 kbps iTunes AAC for encoding
and sometimes Vorbis -q 6 (bcoz ppl in this forum said q 6 don't use intensity stereo)

QuantunKnot,
I know have tuned some Vorbis encoder, do Vorbis use intensity stereo for 128kbps? Is it true that only -q 6 and above use lossless stereo

thanks

I personally find -q 5 more than adequate for playback on my iRiver H120. I feel like I'm wasting bits if I use more since I can't ABX -q 5.

Yes, Vorbis uses lossy "intensity stereo" (in Vorbis terminology, it is known as point stereo) at q 5 and below. At q 6 and above, all stereo is losslessy coupled (no loss of stereo information).

Intensity stereo doesn't seem to be a relevant term since the current Vorbis encoders use either lossless coupling or a mix of lossless and mono. Perhaps you could call it Vorbis' "joint stereo". However, the general stereo model of Vorbis is intensity stereo-like (code one channel, L, and also the difference between two, L-R) and the current encoder uses and mixes both extremes (lossless or mono, ie. L-R=0). Vorbis does not use mid-side stereo. Whereas, to my understanding, intensity stereo in mp3 terminology, is mostly a lossy process. The trick is that Vorbis can select coupling modes on a frequency basis (which mp3 cant, IIRC) and the human ear cannot perceive easily phase differences above a certain frequency. Hence a bit of point stereo isn't necessarily bad, certainly not like intensity stereo, where you either have none or all. Most people generally cannot hear the stereo collapse that easily, so q 5.99 and q 6 should sound, perceptually, identical. It also explains why Vorbis sounds very mono like at low bitrates, where the % of point stereo starts becoming significant in lower frequencies.

Thanks for the replies all. As an ogg newbie it was interesting to compare what others are using.

After a fair bit of listening I've found that Q4 to Q6 seems to be the right range for casual listening.

I basically just put on the headphones and started working my way upward from -q-1 (minus one). In the really low end every q increment gives a drastically noticable improvement in sound quality. After about -q4 though I found the quality improvement for each q increment getting much harder to notice.

-Q4 definitely seems like a good compromise for computer listening if filesize is important, maybe even lower for mobile devices. Eventually I decided that -q5 was the level that was going to be good enough for me, but paranoia is forcing me to go to "enough+1" so I'm currently encoding at -q6.

At the moment I've got both the original ape files and the -q6 converted oggs in my foobar2000 playlist and unless I right-click properties to see the file extension then I've got absolutly no idea which version is playing.

at -q 4, I cannot hear any stereo collapse at all,

but to be on the safe side, I would like to use lossless stereo in case the encoder do miscalculate, is it possible that Vorbis will miscalculate like those mp3 joint-stereo

It would be nice if there is a switch to force lossless joint-stereo, I don't think the quality will degrade much even if lossless joint-stereo is forced at -q 4

Is there switch? i don't see any in the current official Vorbis, maybe when you release another your self-tuned Vorbis, is it possible to add a force lossless stereo switch?

thanks again

for the general/casual user q4.25 will be fine for majority of the available music. q5-q6 for classical and killer samples

Forcing lossless stereo would degrade the quality; perhaps not much, but perhaps a considerable amount, who knows? If there was a perceptual difference it would likely be bad. Why mess with the defaults? They were created by people who know far more about Vorbis technology then you do. What if the encoder was to miscalculate and create some other perceptual error? Why is stereo mode a higher concern for you then pre-echo or HF-boost? Perhaps lossless is the only thing that will satisfy your paranoia.

... just like it's the case in mp3 ...

Beware! You're thinking of normalized residue vectors. After multiplication with the floor curve in the decoder L-R won't be equal to 0 most of the time (in case of simple point stereo) - So I wouldn't call it 'mono' - The intensity (floor curve) can be different and isn't usually the same for both channels.

... same for Vorbis ...
BTW: I consider the lack of something similar to M/S matrixing a design flaw.

An mp3 encoder can dynamically choose a split point in the frequency domain (based on scale factoctor bands). Everything below can be encoded using L/R or M/S, everything else with intensity stereo. It's not that flexible like in Vorbis (the format as such), but the current Vorbis encoder doesn't adaptivly choose a split point (it's hard coded depending on the q level) nor it switches back and forth, so the current usage is pretty equivalent to what mp3 is capable of.

I did a little personal experiment: I could not tell any difference compairing phase shifted noise above 2 kHz and I never experienced any lack of stereo information due to the usage of point stereo. I guess I don't have a problem sample for this case. Can you suggest any test samples ? - Thanks.


SebastianG

*smacks forehead* Good point. I never thought of that. I think it's a result of my previous erroneous understanding that the channels share the common floor curve, which isn't the case, as you pointed out before.



Doesn't Vorbis have lossless stereo coupling which uses polar mapping? The polar mapping seems very intensity stereo-like in nature.



Really? I was reading about the mp3 limitations on http://www.mp3-tech.org/content/?Mp3%20Limitations and noticed this passage:



So how does it actually work?



The stereo collapse at low q's has been reported a few times in the forum. Personally I haven't done any tests but I'm sure there are others around here who have and are sensitive to these problems.

Thanks for the enlightening info. So much to learn, so little time....

My current HDD space (56GB) is not sufficient for Lossless,
I've already installed 2 OS, Linux and WinXP, Linux use up quite
a lot of space......besides I use only NTFS to store my songs, not in EXT3
The highest setting I can go is 320 kbps



I donno why?, at 80kbps, AAC does produce more echo and ringing than Vorbis
But I prefer AAC, bcoz AAC has correct stereo separation
HF-Boost at low bitrate helps a lot!
I still prefer Vorbis over He-AAC, HE-AAC sound muffle to me
this is my preference from first to last

1.highest cutoff possible
2.correct stereo with not too bad pre-echo
3.no pre-echo
4.no HF boost




I'm pretty irritated by the stereo collapse at low bitrates,
especially those songs which has serious spacial effects

Well, I donno how to describe the instrument, there's an
instrument in drumset which produce high frequency sound?
Its definitely not the drum
Its the 2 flat copper colored metal thing that produces
"chhhh, chhhh" sound

Vorbis encode all those "chhh, chhhh" which should be on one side
of the channel to centre channel

For those low frequencies produced by drums are just fine

The conclution that I can made from it is Vorbis save
all high frequencies in mono for lower Q bitrate

I can hear no phase difference at -q 4, but I'm just worried if
Vorbis miscalculate, if someone can just tell me Vorbis
won't fail in calculating lossy stereo like mp3 joint-stereo,
then I will definitely have no doubts anymore

cymbals you mean?

Unfortunately this mapping serves no purpose. M/S matrixing is meant to decorrelate the channels. Square polar mapping produces a dependant pair of values (thus not decorrelating).
Interchannel redundancy is solely exploited by channel-interleaved VQ coding . Since we already do channel-interleaved VQ we could eliminate this explicit mapping by using modified code books. ... I really wonder what Monty was thinking. ... 5.1 coupling requires really large/high-dimensionl codebooks this way.


This refers to M/S stereo (non-IS-coded part) only. You can use L/R for the whole band, M/S for the whole band, L/R (for lower frequencies) + Intensity (for upper freqs) or M/S (for lower freqs) + Intensity (for upper freqs). The split point between lower and upper frequencies for the latter two modes can be selected on a scalefactor band basis (with some limitations).


Thanks, I guess I'll do a search for 'stereo collapse' to find some problem samples then


the high frequency content won't be mono. You'll just get a relative phase difference of 0° compairing high frequency content on both channels. This is believed to be an inaudible simplification. Usually you can't (well at least I can't) tell a difference on most samples.

You could clear your doubts with ABXing.

SebastianG

ABX b4, but no difference, I'm just worried of stereo killer samples come out from no where!

yeah yeah, cymbals,
as the q go lower and lower, the cymbals becomes more and more concentrated in the centre channel

I´d use -q4. Very decent quality (hard to ABX from lossless to me), small size. Ultimately, it depends of your taste/ears/equipment/HD space/etc, but I´m perfectly happy with -q4 for general playback.
Obviously I must confess, as most HA´ers, I go higher when encoding my favorite music. Generally -q6. Even if my ears/equipment don´t notice a difference, I know some day they might.

After doing a fair bit of listening at q4, q5 and q6 I've pretty much decided that I cant tell the difference when listening on my computer, even with headphones. So I think I'll encode all my stuff in q4.

I've archived all my stuff to lossless money audio (ape) so it doesn't matter if I ever need a higher "q" in the future I can always re-encode.

Stereo collapse is one of Vorbis's areas for improvement. Vorbis often collapses the stereo image too much at moderate bitrates. By the time you get up to 90-100kbps, you should not have these collapse issues but unfortunately Vorbis does. And like kotrtim said, it's the worst with percussion.

One of my favorite test clips for this is "Live - Throwing Copper - 01 - The Dam At Otter Creek". From 3:30 on, the uncompressed track has a bunch of crashing cymbals that sound almost out-of-phase, i.e. definitely not from the center but to both sides yet not perfectly solid, kinda weird. But Vorbis just tosses all these high-frequency cymbals in the center (although mid and low frequencies ARE properly dispersed) and this section of the track therefore sounds completely different, very much closed in and claustrophobic. It's off-putting...and this happens at -q 3!

Generally by -q 4 the collapse issues are much smaller in prevalence but they are still noticeable with headphones, if you concentrate on them.

But for "general listening" -q 4 is a good choice...or maybe 4.99, just short of 5 (where the price/performance ratio really starts to tail off). I would err on the high side if you plan to listen on headphones, less if you listen over loudspeakers.

This thread was an enjoyable read. It's interesting to hear what other folks are using. I use Monkey's Audio for everything, and Vorbis was never really a priority for me until I got my Karma, after which I did quite a bit of ABX testing. I found for my ears (with good heaphones on a PC) all samples became 100% transparent at -q6.5, even the most difficult ones (really had to strain my hearing past -q5.65).

However I've been using -q4.5 for quite a while now as it sounds fine on the Karma. I'm currently in the process of moving down to -q4 and I don't think I'll be able to go any lower than that. On my PC I can tell I'm listening to compressed music, but I'm hoping that on the Karma it won't be as noticeable. Even if it turns out that it is slightly noticeable, I'm so far finding the artifacting much less annoying than I do with mp3. Listening to newly encoded classical stuff right now in fact, and it sounds not too bad now that I'm slowly getting used to it. If Rio ever comes out with a new Karma that has a monster hard drive though, I'll probably move to -q6 just because I could, hehe. Sure hope they do some day.

When using vorbis I use q4.25 for casual and portable use.

offtopic: as a matter of fact I'm not using vorbis that much because it drains too much battery from my portable (karma) when compared to good old mp3. If anyone cares, for portable I am using --preset medium with excellent results.

This has been a very good post indeed...



... and has caused me to rethink my current lossy strategy. I had only recently decided to move from LAME -aps to FLAC (now having a larger hard disk) & Ogg Vorbis -q5 for my portable (H120). However, I didn't realise how much battery life Ogg Vorbis files utilise (or the decoder does... right?). Shame, cos I really like the Vorbis encoder and the results are great. Anyway, now I'm messing about with LAME again and which setting to use (not really a topic to discuss on the Ogg Vorbis fourm though eh?).

BTW I've been a keen HA forum browser for about a year now and find it sooo useful. It really is the best Audio site but you all know that anyway

That's the thing that irritates me as well, but both WMA and LAME result in a popping noise between tracks on my Karma. Only FLAC and Vorbis play truly gapless without any issues (haven't tested WAV's). It doesn't really seem to matter what Q level the Vorbis files are at as well, they all seem to drain the battery at an equal rate from what I've seen with my limited testing. The only way we're ever going to be able to have our cake and eat it too is with Musepack support, and don't hold your breath on that one.

Edit: I take some of that back! Different Q values do indeed affect battery life differently. Just tried q9.35 (384k) on a whim and that one album I used drained the battery very very fast, lol. I noticed the hard drive was running non-stop too, which probably explains why. I've also decided not to use q4 (128k) after all, as the artifacting is just too obvious for my tastes (even when I'm not listening for them).

I did further testing. The results were ... unexpected (at least for me)
For those who are interested:

Introduction:
Some people are mentioning point-stereo coded parts sound like they came from the center although different intensity levels are retained for each channel. This implies that we are somehow sensible to relative phase. Indeed our brain correlates the input on both ears and suceeds in determining a delay to estimate the source direction of the sound. (sound travels at a speed of approx 320 meters per second and thus may arrive at different times on each ear).

Test-Sound:
Since Vorbis seems to turn off point stereo for tonal like signals and white noise sounds pretty much like cymbals (a problem that has been mentioned) I produced a problem samples via generating white noise (at 44,1 kHz) on the left channel and copied this noise to the right channel introducing a delay of 15 samples (same intensity!). Due to this delay we get the impression: The sound comes from the left (since it reaches the right ear with some small delay).

What happened:
Some of you may know this already: A delay is an LTI System (linear & time-invariant) and as such it can be characterized with a transfer functio H(z)=z^{-k} where k is the delay in samples. Obviously |H(z)|=1 for |z|=1 which means that the system is an all-pass filter. It only changes the phase of frequencies (depending linearly on the frequency). So, when encoding with point-stereo, the phase differences between left and right are 0. Now, if the split point between lossless stereo encoding and this lossy point-stereo is too low, it sounds like the high frequencies are coming from the center.
Surprisignly I succeeded in ABXing this sample at q5 (!!!) against its original while concentrating on the stereo image. (It was quite hard, but I got a score of 13/15)

I can only speak for myself: I'm surprised about that...
'cause at q5 the band 0-10 kHz is coded losslessly (lossless stereo stereo).

But I guess... for general listening q5 is quite okay.
(I use q4 fo my portable and it sounds good enough)

SebastianG

That's weird that vorbis takes so much more power than mp3, when I listen to them on my computer the cpu utilization is not much different when decoding vorbis as compared with VBR mp3's at similar quality level. Maybe the mp3 decoder has just been more highly optimized for low power consumption on that hardware.

Extract from Gizmodo:
"...
The Karma plays OGG, though it's still a resource hog - you get about 25% less battery life - about 11-12 hours compared to 15+ for MP3 due to the extra cycles and memory requirements when compared to the more svelte codecs. We didn't do a lot of optimisation, so it's running the Vorbis-supplied tremor decoder with only a few tweaks.
..."

Not sure what they mean by "svelte codecs"!?

25% less battery life is about what I've seen with my portable when using -q5.

I've had a look over at the rockbox and they say that for the irver H120, the "...codec decoding is performed on the ColdFire CPU..." (see here)

So, does this mean that it is the physical coding structure of Ogg Vorbis encoded files that results in more system resources/power ("extra cycles and memory requirements") being utilised? Maybe this is indicative of the sound quality / file size trade-off that Ogg Vorbis files has over mp3s?

jb

That's weird

I don't have this problem with my karma player. Everything plays gaplessly...

Quote from the Vorbis docs



Other than the increased memory requirement which we all know about, the former statement, about Vorbis decoding being more computationally simpler than mp3, does seem a bit fishy

The only logical explanation is that the Tremor decoder is far from being optimised and more work needs to be done in that area. Monty slapped Tremor together pretty quickly, AFAIK, so he probably did little, if any, decent optimisation.

Is it possible to just focus on one particular artifact (stereo image) and disregard other (probably more noticeable) artifacts? Given that the source is white noise, Vorbis may produce very noticeable artifacts that may influence your ABXing subconsciously.

Maybe you could try a sample of music that is near transparent with Vorbis at q 5, convert to mono, shift one channel by a few samples, and try ABXing?

Yeah, I thought about that, too. I'm not sure if this proves anything concerning point stereo. (still waiting for ppl bashing my white noise test)

Maybe one can isolate this point stereo effect (without quantization noise) and ABX again...


SebastianG

I would think a higher q value would use less resources when decoding??

Hard drive access eats batteries faster then CPU cycles. FLAC is probably the worst. Actually, come ot think of it, WAV (PCM that is) is probably the worst on those portables that play it.

edit: misquote changed jaybeee @ Dec 7 2004, 03:16 PM ----> Digisurfer @ Dec 7 2004, 10:46 PM

What Gizmodo says about the Karma is bull. Here's a quote from a firmware developer of the Karma:
"Due to efficiency gains in the decoder, current firmware versions should experience less "Vorbis penalty" than the first firmware release did. I don't think we've got figures on exactly how low the penalty is now."


http://forums-riovolution.com/index.php?sh...3&hl=efficiency

ChanFest - cheers, thanks for clarifying. Just shows that there's a lot of sh!te out there on the net.

BTW assassin - in your post you've replied with my name as having quoted something that was actually written by Digisurfer. This then caused music_man_mpc to do the same. Just letting you know so that when you quote people you do it right next time, as nobody likes being mis-quoted.

EDIT: Sorry, forgot to thank QuantumKnot for supplying me with the link to the Vorbis docs earlier in the thread. Cheers mate.

Absolutely. When I copy MP3s to my Zen Xtra, the Li battery drains quickly. I would say that the battery can power the HDD for maybe 1 hour continuously...this kind of activity occurs if you are transferring gigabytes upon gigabytes of data.

However, on playback, the battery lasts 8, 10, 12+ hours per charge. That's because, of course, the hard drive is only accessed to fill the players internal memory buffer and spends most of its time unpowered. The lower the bitrate, the more music you can stuff in that buffer. And MP3 decoding is not very CPU-intensive.

Thank goodness for Gabriel's recent LAME VBR work. I've created my own custom commandline that gives me 124kbps on average and I am pleasantly surprised with the encoder's performance at that moderate bitrate on my portable. LAME won't be able to keep up with today's AAC encoders but it is helping keep the extremely popular and supported MP3 format viable for even audio enthusiasts like ourselves.

Why not use -V5? Its bitrate is only marginally higher than that.

Marginally higher still adds up with enough files. Besides, if he is happy with his command line (and it may sound perfectly fine to him on his particular brand of player) then why waste the space? Especially if one can't hear a difference between the two, then it truly makes no sense at all.

LAME's VBR presets are "general purpose", kinda "one size fits all". But I'm looking to squeeze the most performance (for MY ears) from the least space and because I am creating files for my own use exclusively, I can get a little creative.

One trick I regularly employ is raising the ATH curve (--athlower [n]). I've found, for myself at least, that if I raise the ATH I can save a lot of space without changing perceptual performance. Well, that's not necessarily true: raising the ATH is increasing the risk for artifacts in theory, but I am not striving for transparency. You can't expect such at <128kbps MP3.

When I listen to my DAP I am usually in a less than ideal environment. When I am outdoors, just the sound of the wind, airplanes flying overhead, cars driving by, dogs barking, etc. create enough background noise to obscure the finest details of a recording. So if you have a bitrate budget, you have to make decisions. Raise the ATH and (hopefully) get rid of the info that you can't perceive anyway.

By raising the ATH, I found I could spend more bits in other areas. Transient performance isn't so great with the default -V 5 so I modified the short block threshold to employ short blocks more often. To me, this was a good tradeoff. In a controlled listening test perhaps my choices would prove worse than the default but again, since I am creating MP3s for portable use my priorities are different.

This is my commandline:
"-V 5 --nsmsfix 2 --athlower -22 --athcurve 4 --athaa-sensitivity 12 --shortthreshold 4.7,35 --maskingadjust 0.4 --maskingadjustshort 0.1 --lowpass 15500"

Note that you have to use a 3.97 alpha build because some of these switches aren't available in 3.96. So this is clearly experimental and not HA-sanctioned.

God help us, the mile-long command lines are coming back into fashion
However, you made me curious, so I think I'll give your line a try later on

Greetings,
sTisTi

Long commandlines aren't desirable and you could argue that they aren't necessary. The VBR presets are like tried-and-true stock engines; I'm basically tacking on a NOS kit, fuel economy and engine life be damned. So deviating from the defaults are only for those who understand the risks.

I've done so now, but Lame 3.97a4 (from Rarewares) seems to be totally screwed up: With your command line as well as the proven -V5 --athaa-sensitivity 1 (I also tried the same with V6), I get horrible ringing and warbling artifacts. I tried it with Pearl Jam's "Daughter". With 3.96.1 and -V5 --athaa-sensitivity 1 it sounds like it should - pretty good. Which alpha build did you use?

BTW, the bitrates were:
Your line: 120
-V5 a1: 124
-V6 a1: 119
-V5 a1 with Lame 3.96.1: 129

I'm also using 3.97a4 from Rarewares. I personally encountered one track (Red House Painters - Coaster - 03 - Katy Song) that sounded bad on my commandline. -V5 a1 fared better. That's likely to happen when you "rearrange bits", unfortunately. What I'm after is getting more improvements than regressions.

I have that Pearl Jam album and I think (from memory) that "Daughter" is one of those tracks that gives LAME fits. Ringing and warbling are usually ATH issues and since I am raising the ATH, you'll see this problem become more apparent. I don't think this is an alpha issue. Try -V5 --athaa-sensitivity 1 --athlower 1, for example.

I'm very satisfing at Q2 and above.

@kotrtim

Have you ever tried with Helix DNA producer(or similar GUI producer)?

Yes, the first half minute or so of "Daughter" was one of the killer samples used to tune earlier versions of Lame IIRC.
I tried your suggestion, and it gives me the same bitrate as with Lame 3.96.1 -V5 --athaa-sensitivity 1. It also sounds much better.
However, the quality regression with 3.97a4 and either -V5 or
V5 --athaa-sensitivity 1 as compared to 3.96.1 is blatantly obvious - it's not a question of careful ABXing, it really screams at you - Just listen to 0:22-0:30 or 2:08-2:26 of "Daughter". It's especially noticable with Eddie Vedder's voice, which sounds horribly garbled. Maybe Gabriel can clarify what happened from 3.96.1 to this alpha build.

EDIT: This discussion seems out of place in the Ogg Vorbis forum, maybe it should be split off to "MP3 General" or "MP3 tech".

Are you sure that your ATH settings makes sense ? The ATH is at least 10 dB higher compaired to the standard settings.

SebastianG

edit: oh crap... just realized it's off-topic !