Slashdot Mirror

First I'd love to cite an extremely good video on this topic https://www.xiph.org/video/vid...

I'll try to distil down the relevant portion here.
Nyquist showed us that a bandwidth limited signal sampled by a discrete time system can be reproduced perfectly using 2n samples per unit time where n is the bandwidth of the signal in hertz.

Perfectly isn't hyperbole here. That is mathematically shown.

The other half of digital audio is the accuracy of measurement of those discrete samples. “Bit depth” or bits. While we can reproduce a signal perfectly with perfect samples there is some noise that is added by imperfect sampling of a signal. This is mathematically identical to tape hiss and can be manipulated to less noticeable frequencies using a technique called dithering.

Digital audio can and does faithfully reproduce the original signal with levels of noise below human perception even at a meager 16 bit depth and 48KHz sampling rate (44.1 is also very popular but 48 allows easier low pass filter design).

The stair-steps don't come out of the audio jack, the signal is reproduced by the imaging circuit.
Fast attacks that fall “in-between” the samples are NOT delayed or lost since, again using Nyquist, the signal can be perfectly reproduced (and this is demonstrated directly in the video).

There is a lot of myth and misunderstanding when it comes to digital audio, and there is a lot of truth too. The loudness wars, as other posters have pointed out, has done more to damage the reputation of digital audio than anything else and there are plenty of examples of compressed (both kinds) audio sounding just terrible. One being too low a data rate combined with a terrible encoder, the other just using a small fraction of the overall dynamic range. Those are real issues but they aren't fundamental to signal reproduction.

Hope that explains some of it!

> You can prove it with Nyquist's theorem.

I'm sorry, but this seems a strange, strange claim. See, ears don't do "digital sampling". They have event triggered responses with local analog processing. Thee's a great deal of temporal information buried inside that auditory signal, especially visible in the zero crossings, and ears *do* detect that. A lot of the signal is being lost when it gets undersampled by most modern digital setups.

You should read this article from someone who has deep insight into how ears and human hearing works: http://xiph.org/~xiphmont/demo...

Ditto.

There is a heck of a lot of variability in the material as well. When I have blind-tested rock or pop tracks, I do a statistically accurate (~50%) job of picking the 24bit vs. 16bit or FLAC vs 256/320k mp3. I.e., I can't tell the difference.

On well-produced classical, jazz, and small-group acoustic, the difference can be night and day. 24 bits gives you a larger dynamic range *that those types of recordings can actually take advantage of* and you also get a lower (often invisible) noise floor. (Fwiw, I have degrees in electrical and computer engineering and have done a goodly bit of study on audio, even if that's not my day job. I love arguing with clueless people, especially about some of the counter-intuitive concepts, such as injecting white-noise into a recording to *reduce* the overall noise through the magic of stochastic random processes.)

I can see why people would like the sound of vinyl, especially with a lot of the hugely compressed and clipped audio produced for the mass market, but digital is the obviously superior format when it is done correctly.

Hmm.. The effective dynamic range of 16 bit audio reaches 120dB in practice. That is the difference between a deserted 'soundproof' room and a sound loud enough to cause hearing damage in seconds. And you require more?

This is good reading: http://xiph.org/~xiphmont/demo...

Yes, browsers have indeed become so complicated. It's not just Mozilla, Google's putting even more resources on Chrome than what Mozilla can afford. A browser is now essentially an operating system (see FirefoxOS) that can do pretty much everything *and* needs to do it in a way that's secure against untrusted code (JS). On top of that, Mozilla is involved in projects that reach beyond just the web, like the Opus audio codec and the Daala video codec that I'm personally involved in (there's many more of course).

Most likely ZorinLynx uses Safari or IExploder, as the files are encoded as ogg, and those two browsers are the only ones not supporting ogg.

They can install a codec plugin though.
http://www.xiph.org/dshow/down...
https://wiki.xiph.org/XiphQT

Most likely ZorinLynx uses Safari or IExploder, as the files are encoded as ogg, and those two browsers are the only ones not supporting ogg.

They can install a codec plugin though.
http://www.xiph.org/dshow/down...
https://wiki.xiph.org/XiphQT

The source video is better quality, but the embedded video widget defaults to a lower quality transcoding for streaming (if you click the "webm 360p" box, you can switch to the original video).

Encoding with libvpx also seems to be kind of tricky and at least I've had trouble with getting block-free VP8 files even at a high bitrate (hey Monthy, hurry up and finish Daala ;) ).

I thought [VP8] was roughly comparable to AVC baseline

You're right, I'm not sure why I said ASP.

You might have been thinking of Theora, which as I understand it is comparable to ASP.

It's all but guaranteed some of the next-gen techniques will already have been patented.

I've got a feeling that that's part of why Red Hat has been having Monty make the Daala demo pages, as a sort of defensive publication to make the techniques part of the public record.

One worry I have is that the internet will become a bit like American broadcast TV.

Before Flash gained AVC support, that's sort of what the Internet already was. There were sites that used QuickTime video with Sorenson Video 3 codec, which was reportedly a prototype of H.264. There were sites that used Flash video with FLV1, which is Sorenson's H.263 codec. And there were sites that used DivX, a brand name for an ASP encoder, where ASP is a couple extensions to H.263.

Even as a disliker of Apple, I hear that their little utility which matches your library up with theirs

If the record label hasn't chosen to make its works available through that utility, too bad.

you can get much higher quality than that old CD had anyway

It's not like you could hear any of that quality. In practice, properly dithered 16/44 is enough to cover the entire painless range of human hearing. Or are you referring to serious mastering errors in the original CD, such as those induced by the loudness race?

You are completely wrong about digital audio. There are no square waves, the sample points are infinitely small in time and are used to reconstruct the original analog sound wave mathematically.

Watch this and learn: http://xiph.org/video/vid2.sht...

But the digitized version doesn't play that tone, it plays a mixture of tones with different phases and amplitudes (the convolution of Sin(x)/x with a pure frequency, and only frequencies that happen to be at integral multiples of the sampling rate. The ear can infer it's not a pure tone.

I don't know what you're going on about, but Nyquist certainly doesn't require a frequency to be at any multiple or fraction of the sampling rate to be reconstructed exactly. For any frequency lower than 1/2 the sampling rate, there is one and only one frequency that fits the sample data.

The ear can infer it's not a pure tone.

All tones coming from a DAC are "pure" analog tones. Sound is analog, and electrical waveforms are analog. Did you read Monty's article? The ear reconstructs "in-between" frequencies from overlapping hair cell responses. What's more, the air impacting the ear drum at oscillating pressures doesn't give a shit whether the original recording was analog or digital. You must be just confused about how sound works.

24bit doesn't provide that. Or to put it another way, 16bit provides that exactly as well as 24bit does.

All 24bit buys your is more range between the quietest possible sound and the loudest possible sound that can be recorded, it doesn't add anything "between" the levels, if that's what you're seeking. In fact, there is nothing "lost between the levels" as some people claim, because the sound output is a mathematical recreation with pretty much infinite possible levels within the sample rate and dynamic range the format allows. http://xiph.org/video/vid2.sht... explains all this extremely well.

24bit (and 32bit float) only matter during the recording and mastering process, where you want as much headroom as possible to prevent clipped samples. Once everything is nice and properly mixed and mastered, even the most dynamic soundtracks and music fit easily within the 16bit.

Monty (of Ogg and Vorbis fame) on 24/192 Music Downloads, and why they make no sense.

Read that, or at least accept for a moment that 24/192 is pointless and that a well-encoded MP3 is audibly indistinguishable from a lossless recording in double blind tests.

The whole "confirmation bias" thing is actually terrible for music. It's what makes audiophiles. Somebody tells you music is better with X audiophile feature, and plays it for you. It sounds better. Probably because it's your friend, or the equipment is obviously really nice, or you're used to listening to 128kbps muzak on your earbuds. Once you start believing, there's no limit to the set-ups that can be ruined by not having X feature. Can you ever listen to music again without a $20,000 system?

Audiophilia is like Scientology. The more you believe in it, the more money you have to spend just to be happy again. Demand scientific proof.

Apparently this link hasn't been posted enough times yet. It addresses both your first question (partially) and your second question (in huge detail).

The video you're comparing to is being treated no better than audio. It's simply that human eyes are much better than human ears, so to give a comparable experience much higher bitrates are needed for video than audio.

What all these linear analyses assume is that hearing is a linear process. If its non linear then these analyses are incorrect.

Monty (of Ogg and Vorbis fame) on 24/192 Music Downloads, and why they make no sense.

Most of the point would be to go from MP3 or AAC to lossless. While a 320 kbps mp3 made today will sound far better than a 128 kbps mp3 made fifteen years ago, it still a lossy algorithm that tries to remove sound most people will miss the least. That doesn't mean it's not gone.

Going from CD quality to 24/96 would be another matter, and not likely to bring much, if any, benefit.

It assumes there is no signal above the cutoff (1/2 the sampling rate). If this assumption is not met ie in the real world, then annoying 'aliases' appear in the sampled signal. To fix this, you have to have a low-pass filter. The low-pass filter, by its nature (physics) has to start cutting out signal well below the theoretical cut0off. So there is inevitable loss of signal well under the cutoff.

Oversampling. This is a solved problem.

It assumes perfect, 100% accurate samples and reconstruction. Instead we have imperfect 16 bit resolution samples and heuristic sampling and playback. Reconstructing a good playback signal is a bit of an art. The main impact is the loss of dynamic range. Engineers are forced to limit the dynamic range of the music to avoid excessive loss of accuracy and/or clipping.

Watch Monty's Digital Show & Tell and you'll see why this is not a problem. At all.

I am not a golden-ears person myself but I have friends who are, and gradually they have convinced me that there is a real loss from 16 bit 44kHz samples versys vinyl.

Do a double-blind listening test with them, using a 16/44.1 recording of the vinyl vs. actual vinyl. I would be very surprised if they can do better than chance at telling them apart.

Apparently this link hasn't been posted enough times yet. It addresses both your first question (partially) and your second question (in huge detail).

The video you're comparing to is being treated no better than audio. It's simply that human eyes are much better than human ears, so to give a comparable experience much higher bitrates are needed for video than audio.

Do any of you guys have ears? If you have heard live music vs an mp3, the loss of audio info is very obvious. Many mp3 files--especially rock music--are horrible. Neil is not in this to make money. He's got plenty. He's passionate about music.

A number of double blind tests show that almost no one are able to hear the difference between properly encoded 320kbps and original, including those that are absolutely convinced that they do. The mind is a beautiful thing.

The main problem with Neil is that he is mixing up different issues. Is overly dynamically compressed music a real problem? Absolutely. But that is the mixing and mastering, not related to format. Are there bad low-bitrate MP3 encodings out there? Absolutely, but with higher bitrate and better encoders being the norm it is a problem going away on its own. Are there any reasons at all to go lossless? there is one; if you want to keep the ability to re-compress to different formats/bitrate, then you can avoid compounding of compression artefacts across multiple generations (sort of like how you shouldn't jpeg a jpeg).

And don't get me started on the various snake oil attempts to describe why higher bitrate and higher samplingrates are needed, actually, just read this: http://people.xiph.org/~xiphmo...

No, we already know it's snake oil. See for example Monty's writeup:
http://people.xiph.org/~xiphmo...

Monty (of Ogg and Vorbis fame) on 24/192 Music Downloads, and why they make no sense.

We "played the politics" a few years ago, there was momentum with at one point chrome saying it was planing to ~remove h.264~ from its browser. But in the end that did not pan out. Firefox ended up supporting h.264, and wikimedia was left with very little video participation by its exclusive support for royalty free formats.

Assuming the point of wikimedia is promote free codecs ( not get free information to people that want to access it ) ... Its still too late to say to Apple .. hey if you don't support webm, you won't be able to view the near zero percentage of wikimedia articles that have video content. But when it comes to h.265 and vp9 or Daala, if Wikimedia was a large video player similar to youtube it could help add its weight behind free future free codecs guenteeing they have a prominante home on the web with an active video community.

we're tired of the constant promotion of second-rate codecs that put ideology ahead of technical concerns.

You say "ideology" as if it were just a difference of opinion. Free, open-source software under a free license cannot use patent-encumbered technology. It doesn't matter how good the technology is if you can't use it.

it's not even clear (from a legal perspective) whether these codecs really are patent-free

Actually, you are mistaken on this point. MPEG-LA spent over a year trying to put together a patent pool with which to extract royalties on VP8, and didn't find anything. Then Google gave them some money to go away, and Google has a legal document saying that MPEG-LA will not sue over VP8.

Now, I'm not a lawyer and this isn't legal advice, but it sure looks clear to me: if MPEG-LA couldn't find anything in a year of looking, and signed a legal release on top of that, then there's no danger that VP8 infringes on an MPEG-LA patent. And even less danger that there is some third party out there with a mysterious unknown patent that could swoop in out of the blue to cause trouble (likely for both MPEG-LA and VP8 if it did happen).

At the same time the fact that these codecs are being pushed opposite the existing MPEG codecs only fractures the market and slows the adoption of new video technologies. We end up with Mozilla and Google flailing around with alternative codecs rather than buckling down and doing what's necessary to secure the rights to use the MPEG codecs in the first place, only finally doing the right thing after they've exhausted every other option. Web browsers should have fully supported H.264 years ago.

This is an interesting claim. As far as I can tell, H.264 has not had its adoption slowed even a little bit by VP8 or VP9... could you provide a reference, please?

Google owns and runs YouTube. Do you think Google should shackle themselves to a technology that they don't own, such that they would have no recourse if the licensing authority were to jack the rates up? I think that the business justification for Google spending $100 million to buy On2 and then just give away the technology was to give YouTube a way out if H.264 became rapaciously expensive.

Just as Vorbis never displaced MP3 or AAC, VP8/VP9 may never displace H.264, but if the threat they pose keeps the H.264 license fees from skyrocketing, then those projects were worth doing from Google's perspective.

And, while you may not care about the free software projects such as Debian, I personally think it's good if completely free projects have video formats they can use.

For that matter, I think it's good if completely commercial projects have free video formats they can use. Remember the brouhaha a few years ago where someone read the license for a new video camera, and it appeared that MPEG-LA was going to demand royalties on any video shot with that camera? MPEG-LA "clarified" its position and said that the legal language of the license doesn't mean what laymen think it means... but my understanding of patent law is that MPEG-LA could decide to impose any license they want on new cameras using H.264. I want a camera that uses nothing but free software internally so that I just don't need to worry that anyone has any sort of legal claim for royalties on video I shot with it.

Opus is a roaring success

And I'm hoping that Daala will be an equally roaring success.

https://xiph.org/daala/

Not quite. It's true that for the majority of western music it performs just as well as AAC and Vorbis, however there are certain classes of audio that it does poorly with, in particular polyphonic music. This is an inherent limitation (steming from the pre/post comb filter), that cannot be overcome in future encoders.

This is not actually "true" now. I remember reading a while back that this was one of the major goals of the 1.1 release, and it looks like they largely met that goal.

Look for the section labeled "Tonality Estimation".
http://xiph.org/~xiphmont/demo/opus/demo3.shtml

The short of it is that they have additional code to detect when there are many tones, and when "we consider the frame to be tonal and increase its bitrate." The samples they have of the page seem to show a 25-50% increase in bitrate when it does detect. So, you could easily use it to transparently encode your music library with the caveat that some samples will encode at a significantly higher bitrate. Really though, unless your library consists of a lot of harpsichord music, you're unlikely to see a real impact from this.

CDs are limited to a 90 db

CD has a ~93 dB theoretical SNR, but noise shaping pushes most of that noise above 16 kHz where the human auditory system isn't so sensitive. In practice, CDs can be mastered with 120 dB of dynamic range in those frequencies where it matters. It appears TigerPlish is referring to a 24-bit processing chain, which reduces the noise that each generation of digital signal processing adds, resulting in a cleaner 20-bit master heading into the 16-bit noise shaper. Monty explains.

Or look at it another way. Imagine a 1-bit format that uses heavy dithering to represent signals using pulse density modulation. How much dynamic range does a 1-bit signal have? If not much, why would Sony have chosen 1-bit PDM at 2.8 MHz for SACD?

LPs are limited to 60 db but oddly I have several LPs with more dynamics than their CD counterpart.

That's because level compression in LP mastering works differently from CD. LP uses RCA Victor's New Orthophonic preemphasis curve, which allows bass to go louder than treble, while CD uses no preemphasis.

But the point is, we're not talking about classical music with a 72 piece orchestra, we're talking about what's on the radio worldwide.

I listen to NPR's classical station, you insensitive clod! :p But seriously, recordings destined for pop radio are mastered with very little dynamic range because they have to be audible over a motor vehicle engine that allows very little dynamic range.

Yeah, Monty's writing on these topics is exceptionally clear. His series on the Daala video codec introduces modern video encoding in a way that's amazingly accessible. Maybe he should write a textbook.

That just looks like a typical 44.1khz response graph. Your desktop may have a 48kHz sound card which gives it more "breathing room" above human hearing. (longer tail above 20kHz)

There's still enough room in a 44.1kHz DSP above 20kHz to transmit data though. (As far as I understand it)

see:
http://xiph.org/~xiphmont/demo/neil-young.html

http://www.xiph.org/video/vid1.shtml

That just looks like a typical 44.1khz response graph. Your desktop may have a 48kHz sound card which gives it more "breathing room" above human hearing. (longer tail above 20kHz)

There's still enough room in a 44.1kHz DSP above 20kHz to transmit data though. (As far as I understand it)

see:
http://xiph.org/~xiphmont/demo/neil-young.html

http://www.xiph.org/video/vid1.shtml

Incorrect. This is actually completely doable I used to work for a company that did it (not malware though). You dont have to be much outside the range, even smartphones mics / speakers et al can do this. You only have to go just past 20kHz

see:
http://xiph.org/~xiphmont/demo/neil-young.html

http://www.xiph.org/video/vid1.shtml

Incorrect. This is actually completely doable I used to work for a company that did it (not malware though). You dont have to be much outside the range, even smartphones mics / speakers et al can do this. You only have to go just past 20kHz

see:
http://xiph.org/~xiphmont/demo/neil-young.html

http://www.xiph.org/video/vid1.shtml

I recommend reading Monty's Daala demos 1, 2, 3 and 4. We're not just building a similar codec, but making radical changes to many fundamental components of a video codec.

I recommend reading Monty's Daala demos 1, 2, 3 and 4. We're not just building a similar codec, but making radical changes to many fundamental components of a video codec.

I recommend reading Monty's Daala demos 1, 2, 3 and 4. We're not just building a similar codec, but making radical changes to many fundamental components of a video codec.

I recommend reading Monty's Daala demos 1, 2, 3 and 4. We're not just building a similar codec, but making radical changes to many fundamental components of a video codec.

Looks interesting and all, but it'll take a lot to convince me it's a serious competitor to H.264.

Huge respect for Chris Montgomery. Instead of using his talents to make a huge pile of money he works so we can have open codecs! I also have watched his video destroying myths about digital audio about ten times, it's great.

Shame Quad Stereo took Ambisonics with it. One of those things I wish I never learned existed since it is unlikely the industry will ever start using it again, despite everything being a computer now and there being no need for expensive receivers and whatnot... just point and click a gui to tell the machine where your speakers are and go.

I'm a golden ear. Xiph talks about how you aren't that sensitive, because if you were, you could hear an incandescent light bulb.

Sometimes the noise floor in my house is in fact set by the incandescent light bulb I'm reading by. It's actually a fairly distracting sound, since it brings to mind a mosquito buzzing around. Back in elementary school hearing screenings, I was a few notches above (and below) what the audiologist thought was humanly possible. I've taken care of my ears, and every concert I attend, I attend with ear plugs. It's kind of a pain, actually - most headphones cut off at 20 Hz, and take the bottom out of a recording; I've found very few sets that'll go far enough to hear everything - I've got one electronica track with a baseline slightly below 20 Hz, and I can only hear it at all with a short list of expensive headphones - Sennheiser HD-280 Pro, Logitech/Ultimate Ears UE6000, and Bowers & Wilkins P5s are the only three I've tried that bring out that extra bonus bassline, presumably there either to be felt (in a club), or to be heard by us golden-ears crowd.

20 Hz is a damn, dirty lie. I don't know if it's just me, or if it's actually pretty common, but 10 Hz is about right. Dog whistle tones don't really contribute much musically, but if you like electronic, dub step, or D&B, 20 Hz is a liability and many headphone makers will embellish heavily about their lower end. Sure, it might be able to produce a 20 Hz tone, but it'll be rolled off by 30 dB. Here's hoping you can fix that with the equalizer

I'm going back to Head-Fi now. :p

Did you miss the part in my little story where I was sad to see all the people in the industry couldn't tell the difference? I find it pretty funny that you're sure recording to tape sounds better first, while rejecting the idea that high res audio matters. If you read a study showing that recording to tape initially was inaudible vs. direct to digital in an ABX test, would you still believe yourself here? The mainstream "it has to be measurable via this test to exist" crowd has a long history of losing to audiophiles. Transistors with flat frequency response but poor harmonic distortion, speakers with flat response but bad impulse handling, and digital with poor clock jitter are all things listeners complained about before they were explained in test results. An ABX test can't prove something is inaudible; it only proves it doesn't seem audible the way the test is setup right now.

That most people don't listen very well, regardless of their career or the skills they claim, is endlessly documented. I can't find the study right now, but I even recall one where it was shown that most people couldn't tell 16 bit digital audio from 15 or 14 bits. I think they tested down to 12 and some people still didn't notice.

This whole area has been beaten up pretty well by 24/192 Music Downloads...and why they make no sense. The problem with that whole article is that it presumes perfect equipment, dithering, and recording practice. What 30 years of crappy sounding CDs have shown us is that none of those things consistently happen, because record companies spew out whatever half-assed reissue garbage they think they can get away with. Perfectly made CD audio sounds excellent, but there are so many examples of less than perfectly made ones.

I suspect that the main reason 24/192 recordings sound better is that it assures a baseline high quality of equipment was used, and there's a very large margin for error while still capturing everything. All sorts of mistakes are absorbed by the medium. Also, asking for a 24/192 recording states a preference--that the consumer wants the audio without the excessive compression or other processing. Trent Reznor was just in the news for offering regular and "audiophile" versions of his new record; that's the same sort of distinction people asking for 24/192 recordings are saying they prefer over the mainstream CD quality version.

The higher the frequency the more aliasing distortion you have. A fifteen kHz tone has only three samples per crest making a sine wave indistinguishable from a sawtooth wave. Tripling or quadrupling the sampling rate would greatly reduce aliasing.

This is totally wrong. Any more than two samples per cycle adds zero additional information. This video may help you see the light.

The tl;dr is 'No'.

The longer explanation from xiph.org

You're quite wrong.

The thing is... 16 bits is enough, but only barely. A quiet room is about 30-40 dB above the threshold of hearing, and 16 bits gets you about 96 dB of signal-to-noise. I think it makes sense to add those numbers, and say that if you set the volume so you can just barely hear the quietest bits of a recording that covers the entire dynamic range, then the loudest parts will be at 126-136 dB. Coincidentally or not, that's actually right at the threshold of pain, which is typically quoted at 130 dB.

OK, so this is not actually quite right:

The answer: Our -96dB noise floor figure is effectively wrong; we're using an inappropriate definition of dynamic range. (6*bits)dB gives us the RMS noise of the entire broadband signal, but each hair cell in the ear is sensitive to only a narrow fraction of the total bandwidth. As each hair cell hears only a fraction of the total noise floor energy, the noise floor at that hair cell will be much lower than the broadband figure of -96dB.

Thus, 16 bit audio can go considerably deeper than 96dB. With use of shaped dither, which moves quantization noise energy into frequencies where it's harder to hear, the effective dynamic range of 16 bit audio reaches 120dB in practice... (source)

So there's rather more headroom (by about 20 dB) than I say.

However, even that link has this to say:

Professionals use 24 bit samples in recording and production [14] for headroom, noise floor, and convenience reasons.

16 bits is enough to span the real hearing range with room to spare. It does not span the entire possible signal range of audio equipment. The primary reason to use 24 bits when recording is to prevent mistakes; rather than being careful to center 16 bit recording-- risking clipping if you guess too high and adding noise if you guess too low-- 24 bits allows an operator to set an approximate level and not worry too much about it. Missing the optimal gain setting by a few bits has no consequences, and effects that dynamically compress the recorded range have a deep floor to work with.

An engineer also requires more than 16 bits during mixing and mastering. Modern work flows may involve literally thousands of effects and operations. The quantization noise and noise floor of a 16 bit sample may be undetectable during playback, but multiplying that noise by a few thousand times eventually becomes noticeable. 24 bits keeps the accumulated noise at a very low level. Once the music is ready to distribute, there's no reason to keep more than 16 bits.

so my overall point -- 16 bits is enough for the end user but not very good for mastering -- holds.

24-bit makes sense, giving far greater dynamic range (which can be construed as resolution if we want to compare it to photos/videos). Admittedly, calling it 24-bit is a bit absurd as the best I've heard of is closer to 20, maybe 21 bit, but if we're trying to keep within a standardized system, may as well use groups of 8. In older recording/playback system 48k was a vast improvement over 44.1k. The perceived advantages to 88.2k, 96k, 176.4k, and 192k were due to a one octave (88.1k/96k) or two octave (176.4k/192k) low pass filter causing less of a high frequency bump than a tenth of an octave (44.1k) or an eighth of an octave (48k). This is not really necessary anymore as the digital filters perform way better than most people give them credit for.

As a playback standard, 24-bit 44.1k or 24-bit 48k make perfect sense with current generation, decent quality D/A. 24-bit permits the greater dynamic range and greater dynamic accuracy that pieces like Chopin's can benefit from. There likely will be an audible sonic difference between 44.1k and 192k, but it will be distortion. Some people certainly prefer the sound of these higher bit rates, however it is still not accurate to the original product. If the higher resolution bit depth isn't necessary (as is the case with most modern music) it will not be detrimental to the playback, unlike 192k.

For anyone looking for a more in depth write up, it was shared here on /. a while back, but there's a great write-up from Neil Young about why these formats don't matter (the argument using solely a 1k test tone is very easy to dither, using a full symphony or even a full piano's range is virtually impossible to mask with dither). I disagree with him in general on the 16-bit vs 24-bit, but, for the most part, the average listener would never know the difference considering the dynamic range in most modern music is still comparable to watching a movie that's 128 x 72 upconverted to 1080p while 1080p would've been available to the producer to begin with.

Short of ogg-vorbis (which is a file format I've known exactly one person who gave a shit about), I have yet to encounter a file format not supported in iTunes. And for all I know it supports ogg-vorbis, but since I don't own or want anything in that format, I don't give a damn.

You can resolve that example too: http://www.xiph.org/quicktime/

Ogg Opus (open, royalty-free, not patent-encumbered audio) beats the pants off of HE-AAC (which, in turn, is superior to everything else at pretty much every level).

Wow! So much wrong in just a single sentence...

Opus is an IETF developed codec, based on CELT from Xiph.org, and Silk from Skype/Microsoft.

Ah, I thought Opus was under Xiph's Ogg umbrella. Xiph certainly hosts the new CELT+Silk project called Opus. Ogg is (afaict) the only container used by Opus (.opus is an Ogg file container). Noting those things, if you can refer to Vorbis as "Ogg Vorbis" then why can't you refer to Opus as "Ogg Opus?" I knew this was informal (though it is somewhat common) and did it mostly to concisely demonstrate its connections to Xiph (which most people know as "the ogg [vorbis] guys").

HE-AAC certainly isn't "superior" at "every level". It excels at very low bitrate encoding that sounds SOMEWHAT like the original. As you start increasing the bitrate (eg 96k), low-complexity AAC easily surpasses HE-AAC. And as you go to higher bitrates still (eg. 160k), temporal domain codecs can outperform any frequency-domain codecs, so Musepack will beat the pants of AAC, and even Opus.

You again caught me generalizing. Perhaps I was mistaken, but I thought AAC encoders capable of HE-AAC would automatically select which AAC codec to use based on the compression level and this is what I was referring to. I also should have said "roughly equal or better than" rather than "superior to" in that statement.

Regarding Musepack (MPC), that is an obscure format that was generally on par with Vorbis back in the day (but Vorbis has been improved a few times since then while Musepack has not). I'm gauging most of the surviving "next-gen mp3" codecs as largely equivalent at the 128+kbps level these days (though my personal preference, biased in part towards Freedom (and Linux compatibility), is for vorbis).

We don't seem to be getting quality listening tests for higher bitrates any more. Everybody is obsessed with super-low bitrate (so much so that they don't even try to find an mp3 baseline by which to state things are roughly equivalent to). I found a 64 kbps comparison which shows Opus narrowly beating Apple's HE-AAC for the lead. I'd love to see a thorough and up-to-date comparison of the major contenders at the 96, 128, and 160 kbps levels that also includes the lossless version as a baseline.

Still, low bitrate lossy audio quality is important, so Opus is a good choice for streaming audio and video. That's why Google chose it for their latest revision of WebM, along with their new VP9 codec that they claim outperforms HEVC.

That's odd, since WebM uses Matroska,which doesn't yet support Opus (though that's in the works). We'll see if Google can successfully make MPEG-LA obsolete. For that, we'd also have to compare VP9 to H.265 (Google noted VP9 was ~7% behind h.265 in Q4 2011) or perhaps wait for Daala or some Dirac-like wavelet codec to come out of left field (though wavelet compression has large hurdles, it is closer to how our eyes actually see).

I seriously doubt the MPEG / MPEG-LA organizations, and their members, will consider using a patent-free audio codec along with their heavily patent-encumbered video codec. Their business model is patents, and they'll chose an expensive and inferior option over a free one, any day. I'd expect HE-AACv2 to be the best you can count on for the foreseeable future.

Agreed. Hopefully that won't stop Xiph and Google from stealing the show.

Ogg Opus (open, royalty-free, not patent-encumbered audio) beats the pants off of HE-AAC (which, in turn, is superior to everything else at pretty much every level).

Wow! So much wrong in just a single sentence...

Opus is an IETF developed codec, based on CELT from Xiph.org, and Silk from Skype/Microsoft.

Ah, I thought Opus was under Xiph's Ogg umbrella. Xiph certainly hosts the new CELT+Silk project called Opus. Ogg is (afaict) the only container used by Opus (.opus is an Ogg file container). Noting those things, if you can refer to Vorbis as "Ogg Vorbis" then why can't you refer to Opus as "Ogg Opus?" I knew this was informal (though it is somewhat common) and did it mostly to concisely demonstrate its connections to Xiph (which most people know as "the ogg [vorbis] guys").

HE-AAC certainly isn't "superior" at "every level". It excels at very low bitrate encoding that sounds SOMEWHAT like the original. As you start increasing the bitrate (eg 96k), low-complexity AAC easily surpasses HE-AAC. And as you go to higher bitrates still (eg. 160k), temporal domain codecs can outperform any frequency-domain codecs, so Musepack will beat the pants of AAC, and even Opus.

You again caught me generalizing. Perhaps I was mistaken, but I thought AAC encoders capable of HE-AAC would automatically select which AAC codec to use based on the compression level and this is what I was referring to. I also should have said "roughly equal or better than" rather than "superior to" in that statement.

Regarding Musepack (MPC), that is an obscure format that was generally on par with Vorbis back in the day (but Vorbis has been improved a few times since then while Musepack has not). I'm gauging most of the surviving "next-gen mp3" codecs as largely equivalent at the 128+kbps level these days (though my personal preference, biased in part towards Freedom (and Linux compatibility), is for vorbis).

We don't seem to be getting quality listening tests for higher bitrates any more. Everybody is obsessed with super-low bitrate (so much so that they don't even try to find an mp3 baseline by which to state things are roughly equivalent to). I found a 64 kbps comparison which shows Opus narrowly beating Apple's HE-AAC for the lead. I'd love to see a thorough and up-to-date comparison of the major contenders at the 96, 128, and 160 kbps levels that also includes the lossless version as a baseline.

Still, low bitrate lossy audio quality is important, so Opus is a good choice for streaming audio and video. That's why Google chose it for their latest revision of WebM, along with their new VP9 codec that they claim outperforms HEVC.

That's odd, since WebM uses Matroska,which doesn't yet support Opus (though that's in the works). We'll see if Google can successfully make MPEG-LA obsolete. For that, we'd also have to compare VP9 to H.265 (Google noted VP9 was ~7% behind h.265 in Q4 2011) or perhaps wait for Daala or some Dirac-like wavelet codec to come out of left field (though wavelet compression has large hurdles, it is closer to how our eyes actually see).

I seriously doubt the MPEG / MPEG-LA organizations, and their members, will consider using a patent-free audio codec along with their heavily patent-encumbered video codec. Their business model is patents, and they'll chose an expensive and inferior option over a free one, any day. I'd expect HE-AACv2 to be the best you can count on for the foreseeable future.

Agreed. Hopefully that won't stop Xiph and Google from stealing the show.

Ogg Opus (open, royalty-free, not patent-encumbered audio) beats the pants off of HE-AAC (which, in turn, is superior to everything else at pretty much every level). Opus also streams better, capable of dealing with extreme low-latency demands associated with real-time uses like VoIP.

It is so common to see people talking about tweaking x264 to improve quality and compression, but there is a point where you're better off optimizing the other pieces; AC3 passthru is laughable contrasted to 6-channel vorbis (which I use in place of HE-AAC due to not having access to a quality AAC encoder on Linux). I'm still waiting for opus support in matroska (which is in progress) or something to supplant matroska as the prevailing file container.

There's also the patent question; will this be intentionally patent-encumbered the way MPEG standards tend to go (in which case they'll certainly connect it to HE-AAC), or will this be a somewhat more open standard (which lends nicely to Opus)?

Tell you something FLAC doesn't have, though - a consistent mapping for 5.1 or more channels.

This has been addressed in FLAC 1.3.0, which came out recently. This is the first update to FLAC in six years, you'd think someone would have posted something about it to Slashdot.

It consists of an inherently lossy encoding in the frequency domain (like MP3) plus an encoding of the difference between the lossily encoded audio and the original.

While a few other lossless formats do this (mostly for backward-compatibility), FLAC does not convert the audio into the frequency domain. It either uses a polynomial or linear function: http://xiph.org/flac/documentation_format_overview.html