Stereo or Joint Stereo? 44100 or 48000?
 

I've been using LAME and I'm wondering if I should do everything in Stereo, Joint Stereo, or if it depends on the soundtrack.

Obviously, 320 Kbps is what I want for best quality.

Also, should I do everything at 44100 Khz or 48000, or will LAME even let me decide? I'm wondering how that factors into quality, if at all.

Responses will be greatly appreciated.

Joint Stereo is what you call it when the same sound comes out of both speakers. It's like monaural multiplied by two. It sucks, go with stereo. With regards to sample rate, CD audio is recorded at a sample rate of 44.1Khz, so going above that will not bring better results unless you have a source which is better than CD audio.

320Kbps is nothing short of a waste of your disk space. If you're going to go that high, you might as well just start using a lossless format anyway.

As much as I realize 320 Kbps really isn't noticable, I haven't heard anybody tell me the point where it IS noticable. Some people claim "I can hear artifacts all the way up to 256 Kpbs!" I could have sworn somebody on these boards set that, actually. So I just go with 320 because then nobody can argue that this is a perfect rip.

I don't like lossless formats, they're a pain in the ass.

And thanks for the info on the stereo thing, especially since I was actually going to go with joint just because it sounded cooler.

KHz I'll just leave alone.

Wow, that's just plain wrong! Think of Joint Stereo as a way to encode the stereo signal losslessly. It fully preserves the stereo channel information! Full Stereo encodes both channels seperately, thus reducing the effective bitrate of the encoded material. So using the same bitrate, a JS encoded file will sound superior to a FS encoded one.

LAME's method of encoding Joint Stereo is superior to Full Stereo with virtually any material. That's why knowledgable people usually suggest using Joint Stereo over Stereo (at least when encoding with LAME) any time.

The way JS does it is to fully encode one channel while the other is just a difference. Since most music will appear to be the same on most recordings for both channels the difference should be negligible hence less data. JS should use more CPU time to decod,e however that's a non issue nowadays.

The LAME authors have already stated that there is no quality hit when using Joint Stereo. In fact, regular Stereo may sound worse then JS because you are encoding both channels seperatly.

Just use LAME 3.97b2 with the commandline "-V 0 --vbrnew". It's the top quality VBR. You won't hear any difference between that and 320kbps cbr.

And you don't need to worry about the sampling rate, just use whatever the original files were.

I beg your pardon. I was going on what I had read, and I can't apologise enough. Since reading your responses, I've found further information that confirms what you are all saying. It does seem to be a popular myth that Joint Stereo is exactly what I said, but it's not apparently true. I've even conducted my own test.

In answer to your other question, I find that with MP3, 192kbs is about as good as it gets. Above that, I'd rather use lossless.

Good to see that's settled. :)

About the sampling rate (kHz). It's best to keep the same sampling rate when encoding. Meaning, if your source is 32kHz, then encode to 32kHz. If it's 48kHz, then encode to 48kHz.

Or you can take compatibility into consideration. In that case you would want to indiscriminately encode everything at 44.1kHz. That way old/crappy MP3 players can play the MP3's.

http://flac.sourceforge.net/documentation.html
Look under 'INTER-CHANNEL DECORRELATION' -> Joint-Stereo is in general a lossy process. It can be lossless, but only when Middle/Side-Stereo encoding is meaned by Joint-Stereo. In case of the the 'intensify stereo' algorithm phase information is lost during the encoding process. If thats audible is another question, but in general joint-stereo is not lossless.

cya
liquid

LAME is lossy by nature as it is a mp3 encoder. The point is that joint stereo is more effective compression and saves bandwith, yielding higher quality audio. Similarly to how ABR 128kbps will sound better then CBR 128kbps, because you are using more effective compression. It's used in the -V system for a reason.

From the lame docs:

I wasn't talking about LAME, wasn't I? I was talking about the term joint-stereo in general. That LAME/(every other mp3 codec) is always lossy should be clear to everyone that listens to music on his computer.

I didn't doubt this. But you should add that efficiency is decreased if the channels don't correlate much (normally they do).

I know the LAME codec and that it is using the Middle/Side-type of joint-stereo. But this process doesn't introduce artifacts (and therefore is completly reversible). What degrades quality is the quantization of the resulting bitstreams (different amount of bits are allocated).

Wow, lots and lots of replies, thanks all.

It's all right for the mistake, Soluzar, I hadn't started yet. Good thing I came back here before I did. Also, my Ephemeral Fantasia rips are all mystically ending up on Joint Stereo by default so I guess that's a good thing.

And since most everything is 44100 KHz anyway, I guess I don't have to worry about that too much. Although come to think of it, Unreal Tournament is, for some strange reason, in 48000 KHz, now that I look at Winamp. But oh well. Thanks for the help in achieving ultimate quality.

Hmm, well, to be honest, I'd expect a lot of VGM to be at 48kHz these days. And older VGM could be anything. The only thing I'd ever really count on being 44.1kHz is CD-audio.

Looks like you're doing good for yourself.

Hey you are ripping Ephemeral Fantasia too????? I ripped it a while ago. Glad to see someone else likes the music as much as I do because virtually noone knows about the game. If you can make a rip in joint stereo I would be interested in checking it out to hear the difference since mine is just in regular stereo.

I use vbr-new with a forced stereo. First, joint-stereo uses a single channel. Have it at you will, but the other channel only "echo's" the single original one. vbr solves all of the problems with file size compression because it naturally tries to find the suitable bitrate for each second of the song.

Thus, stereo remains higher quality in this case. Also, the truth is, if you use any mixing or effects on your sound card (usually higher quality ones), stereo will end up sounding better because it keeps the original CD's data the same (except compression).

Wrong.

Also wrong.

Well, no. Did you even read the thread? Joint stereo does not use a single channel and seperate stereo is a waste of bandwidth (at least with Lame).

This forum has a Private Message system. Use it.

192k is certainly not as good as it gets. I had a rip of a score in 192k and played it almost every day for a long time. I got used to it and thought it sounded great. I then got the CD and did a re-rip using API. WOW. What a difference! A much better, clearly, fuller sound with more high-end and less artefacts.

192k still cuts off a lot of high end. If it was "as good as it gets", higher bitrate encoding wouldnt have been invented! :)

This could be for many reasons. Rips that you don't do yourself (or that don't have files documenting the settings of the rip) aren't worth basing conclusions off of because:

1) it could be using an old, crappy quality encoder. Blade or Xing, anyone? Good thing mostly no one uses these things anymore.
2) some people got the brilliant idea to re-encode 128Kbps MP3s in a higher bitrate (rather than encoding from the original CD audio), erroneously thinking that it would increase the quality.

Also, I would warn that the psychological effects of seeing -- or just knowing -- that one file has a higher bitrate can influence your opinion about which sounds better. Many people claim to be able to hear a difference between MP3s of different high bitrates, but when they are played audio without knowing the bitrates of each they suddenly are unable to perceive any audible difference between the two.

Moguta,

Great avatar by the way!

I hear what you say, but the basic fact is 192 still cuts off a lot of frequencies, whatever encoder you use.

As for the psychological effects, sure, I guess it happens. But when I did that 320k rip of that score, I knew it sounded better without hearing the old 192k rip. Instantly, I was impressed by its quality. It wasnt until I had played a few tracks that I went back to compare a couple and the differences were apparent.

When you encode to 192kbps with LAME, it uses a lowpass filter of 19.5kHz. Most CD's are mastered with a lowpass at 20.0kHz. I wouldn't call a measly 500Hz a lot. Besides, you can choose to raise the lowpass if you want to - you will probably only be slightly diminishing the quality of the MP3.

Did you make the "old 192k rip" yourself?

But you are assuming all 192k rips have data up to 20Khz. They dont. A lot of dance music I ripped in 192k that was complex with a lot of varying frequencies was cut off at 16khz. It varies, depending on the complexity of the material.

Yes, I did. Again, there wasnt much high-end above 16khz or so.

You're right. The more complex the material, the more distortion and less frequency content you will get at the same bitrate.

Bitrate needs vary depending on the complexity of the material. Which is why VBR should be used rather than CBR.

I just use 320k for everything.

And most of the time, it's a waste.

If you want perfection, don't even use MP3.

If you want the best tradeoff between compression and quality, use VBR.

Most? Not all?

I didnt say I wanted perfection.

VBR takes longer to rip and encode and you get VERY SLIGHTLY better sound with 320k. Its quicker for me to rip in that. Thats the main reason I do it, actually.

That's nonsense. Ripping has nothing to do with the encoding process.

I doubt that, any proof? Today's optimizations on the LAME codec are mostly done in the VBR area, so IF VBR encoding does take more time than CBR encoding its about a few seconds (or even ms).

It only does sound better if the VBR estimation algo is bad and allocates too few bits for the frame. Of course 320kbit/s CBR is the best encoding mode you have, because the peak bitrate of a frame can't exceed 320kbit/s - so that's the final limit.
BUT CBR always allocates the full bits for one frame, even if the audio data can be reconstructed using fewer bits. So VBR gets you the same quality (IF the estimation algorithm is not flawed, what we assume) but consuming fewer space.

True, my mistake. But encoding does take longer. I did my own test and it took longer.


I know the deal. I just prefer to give every frame the maximum limit. How can ANYTHING but the most basic sound information need less than 320k? I dont care if the files are bigger.

Most. Some frames need 320kbps. Some need even more but 320kbps is the most they can be given.

You only have to encode once.

Well as Ive said before, who's to say anything needs less than 320k (unless its silence or really basic material)? So what if a few seconds takes up X-100k as opposed to being a few K of disc space? I want every frame to have the best quality available, even if the encoder programmer/code-writer/someone else says its not needed. Its subjective.

I rip from CD using dBpowerAmp. It rips and encodes in one single step. I dont rip to WAV and then encode to MP3. I do it all at once. But its six and two threes, the point is that encoding to APX takes me longer. Its fact, I tested it myself :)

Guys, this argument really isn't necessary, and it's getting a little too fanatic. MP3 encoding is not about absolutes, its all approximation.

ArrowHead, of course encoding in VBR takes longer than CBR. The greatest power of a good VBR algorithm is that, for each frame, it tries to determine the best number of bits to represent that short segment with audible accuracy. And that determination is probably somewhat calculation-heavy, taking additional processor time. CBR simply doesn't need such a step since it already knows how many bits will be used. It's also Iceboy's choice whether he wishes to sacrifice disk space for quicker encode times, despite the fact that you only encode audio once. Maybe he has a whole lot to rip!

And Iceboy, I just want to correct what you're stating... most audio does not in fact "need" anywhere near 320Kbps to sound exactly the same as the original, to the typical human ear. That last bit is most critical: to the typical human ear. Even 320Kbps MP3s remove many frequencies from the audio, based on the idiosyncrasies of human hearing (known collectively as 'psychoacoustics') : one simple property being that a louder, lower frequency can make a simultaneously-played quieter, higher frequency completely inaudible!

The fact is that much of a waveform is completely unnecessary when your only concern is human perception, and lossy codecs take advantage of this. Over in the Hydrogen Audio forums, where many digital audio enthusiasts, audiophiles, and codec developers reside, one recurring theme highlights the over-sufficiency of 320Kbps. Whenever folks have tried to organize a community ABX-double-blind test comparing each format's best encoders at any bitrate higher than 128Kbps, they have a very hard time finding a significant number of people who can -- even under isolated listening conditions and repeated, focused listens -- verifiably rate the codecs against each other. There have been many instances of testers ranking the original audio below some of the compressed audio! 320Kbps is far from necessary to achieve audio that is no perceptibly different.

However, because the human hearing system is so complex and organic & because lossy audio encoders must use simpler algorithms that balance computing power/time vs. quality, certain so-called "killer" samples will cause evident flaws even at 320Kbps. Fortunately, such samples are usually uncommon. Their failure is usually less related to the bitrate used, and more related to the MP3 encoder not being 'tuned' to properly handle certain occurrences in audio. But it is certainly difficult to try to take into account the infinite number of ways audio can take shape, and LAME developers are continually refining the encoder against the latest "killer" samples. But yet it's still a sign that if you want to assure your music will always sound exactly the same as the original, you have to go with some totally lossless form of compression, such as FLAC (my preference) or Monkey's Audio. The disadvantage of this, however, is that while 320Kbps compresses to approximately 22.7% of the size of a CD-quality WAV, lossless codecs tend to range from 25-80%, typically more toward 60% or so.

Iceboy, it is entirely up to you whether you use 320Kbps MP3 or not; I cannot take issue with your choice. My only wish was to clarify your misperceptions, for you and and others who read this thread. I do wonder, though, if LAME's ABR mode encodes more quickly than the VBR presets, since I know encoding time was an issue for you...

Additional Post:
The thing is these effects are not consciously felt, they are very subtle and subconscious. You can easily "hear" the 320Kbps songs as higher quality, because in your mind you know that 320Kbps is the highest setting possible, thus it ought to sound better. And as soon as sound hits your ear and passes, all you have left is a cursory mental impression, so you can't objectively take the sound and compare the full details side-by-side in your mind. Do more exploration into the subconscious power of mental suggestion... documented studies have been done on some really freaky effects. Nothing like ESP or superstitions or anything like that but... well, I'm kind of getting off-topic, so here's one interesting read on one aspect of the matter: http://www.quackwatch.org/01Quackery...ideomotor.html

It's also possible that those 192Kbps files -were- a bit crappy. I don't know if you said you encoded those with a recommended LAME encoder and settings, but LAME recently hasn't focused on CBR quality anyway, instead implementing good ABR and VBR and emphasizing their advantages. But I'd, for fun, challenge you to double-blind test yourself between those 192Kbps and 320Kbps samples with foobar2000's built-in ABX comparator (select two tracks -> right click -> Utils -> ABX two tracks). It's proven to me how easy it is to casually "hear" an improvement that, under examination, isn't really there. It's rather interesting.

Well unless I see conclusive proof, I cant believe this.

Whats your problem with not believing me? It stands to reason a complex piece of 320k orchestral/electronic music will sound better than a 192k rip, because there is more data to play with! For one, the 192k cut off everything above 16Khz or so, whilst the 320k kept everything up to 20khz. The added high end was noticable to me right away!

You know, it is possible for some people to hear differences. Going from what you have said, you dont seem to believe that anything above 192k sounds better. Some people have very sensitive ears. Im one of those people.

I cant do those double blind tests, because I naturally got rid of the 192k and replaced them with the new 320k. But if I ever get the chance again I will compare some files, just for the hell of it.

Well believe it or not, Ive been a practising magician for 16 years now, and I know a lot about suggestion and all that :)

So rip and encode again. The only way you're going to get your proof is with a double-blind. If you don't have the original anymore (or you're too busy or something), I can select a track from my CD collection and prepare your test files for you. Take about 5 or 10 minutes, plus transfer time.

In the meantime, please drop this. It's been explained several times now why "more data to play with" is irrelevant--it's just not getting through to you I guess. I realize that you don't understand, and I wish there was a way that we could say it differently so that you would get it, but I'd hate for you to give someone else the wrong idea about all of this just because you haven't heard the proof.

At the end of the day, 320k gives ever so slightly better sound at the expensive of bigger files. I can live with that.

The only thing I cannot believe unless I see (not hear) some sort of statistical data is how ANYTHING but the most basic music can need less than 320k in order to make a sound as close to the original as possible, using VBR. Thats all.

Funny thing is that orchestral and electronic music does compress best in general. FLAC encodes of orchestra music show this very clearly.

The more 'noisy' music gets the harder it is to compress (because of entropy).

Not beint funny but how do you mean? Ive never used FLAC, thats all.

What do you mean?

But that's not always the case.

When VBR is used, 192k is great as an average. When higher bitrates like 320 are needed, they're used.

When using VBR, high bitrates are used when they're needed and lower bitrates are used when they're not. So you can tell that a certain piece of music "compresses well" if the average bitrate comes out pretty low.

FLAC is lossless compression of wave data (I hope you know that). And FLAC is therefore VBR by concept.
Method is (simplified) like that: Waveform is analyzed and approximated using some mathematical functions. The parameters can easily and efficiently be stored and even compressed further. But this approximation is not perfect, so there is some residual error which is compressed and stored in the bitstream.

This means that the more 'order' music contains ('order' in the sense that the waveform is easily approximated using mathematical functions - for details see the FLAC homepage) the better it is compressed. If you go from 'ordered' music to 'noisy' music compression gets worse. Reason is simple: You can't compress noise - that's because of entropy.



By entropy? If you're not familiar with the concept of entropy (information theory) then you should read a wikipedia article or something similar.

Additional Spam:
I just wanted to note that FLAC encodings are a strong indication of how good the music data actually compresses (compression = lossless compression).

Thanks for your info, that was quite interesting.

What we mean by more entropy or more noise is essentially audio with less of a pattern. Compression -- whether MP3, DivX, or ZIP -- utilizes the presence of predictable patterns to, in essence, remove redundancy and save space. However, as data approaches the purely random (entropy/chaos/"noise"), it gets more difficult to compress.

As for statistical data to show that anything less than 320Kbps is necessary... well, this isn't quite that, but it's a 128Kbps listening test conducted almost 3 years ago.
Test Introduction
Test Results
The participants are probably all self-professed audiophiles, since it was conducted on the HydrogenAudio.org forum. Double-blind tests were conducted with those formats vs. the original audio, and people were asked to rate the quality subjectively from 1.0 to 5.0, 5.0 being perfectly indistinguishable. Note how close some of these codecs are rated to 5.0 at just 128Kbps! So, it's certainly not improbable to think that bitrates above that could reach the level of being audibly indistinguishable. Although, I admit it would be wrong to decisively conclude that just from the results of this test.