Speculation On a Lossless iTunes Store

DrJenny writes "C|net UK has up an interesting blog post predicting that within 12 months Apple's iTunes Store will include a download center for lossless audio. This would be a massively positive move for people who spend thousands of dollars on hi-fi gear, but refuse to give money to stores that only offer compressed music — they could finally take advantage of legal digital downloads. The article goes into details on how Apple's home-grown ALAC lossless encoding relates to FLAC, DRM, and the iPod ecosystem."

"Lossless"? Such BS

[timster]

Forget "lossless" when you've already lost so much of the original wave by mixing it down to 16-bit 44khz stereo in the first place. I'd rather have something that started out with a higher sampling rate/etc, but with good lossy compression to pull it down to something that doesn't require DVD-type storage for a single album.

Let me translate...

[Anita Coney]

From the blog:

"And now I have an inkling Apple will add lossless music downloads to the iTunes Store within the next 12 months."

Translation:

I have no fricken clue that this will ever happen, but because I think it'd be cool if it did, I'll go ahead and blog about it.

So...

[Seakip18]

This would be a massively positive move for people who spend thousands of dollars on hi-fi gear, but refuse to give money to stores that only offer compressed music

So....this is for all 17 of them? I figured since they have that much money for equipment, most would just get the CD's and rip them via those means. If you can afford a $20k speaker, you can afford a a few TB Hard drives to keep your music.

Finally! DRM free, lossless, digital music!

[Lord Byron II]

Isn't it amazing that 25 years after the release of the CD, we're excited to finally have a way to buy DRM free, lossless, digital music? If this happens, we'll be back inline with 1982 technology.

Re:Or they could just stick with CDs

[Vegeta99]

You might like all the songs. Since I listen mostly to rock/alternative genres, I probably only want his singles. Can I buy them cheaply, like my dad could in the 70's? He's got a stack of 45's that probably reaches the ceiling of my apartment.

I do realize they still sell them, but are they $0.99 per song cheap?

DRM silliness

[spiritraveller]

So they lock down these files with DRM. Then DVD-Jon (or someone else) comes up with a DRM-stripping program for the files.

Then people can re-encode the files to their format of choice. But by then, most consumers have said "fuck it" and decided to just download their format of choice directly from p2p or usenet because it's easier and simpler than paying Apple and still violating the DMCA just so the music they paid for will work on the audio player they own.

Oh wait, that's already the status quo... Never mind.

Cool if/when it happens

[Strange Ranger]

Sure I might buy something in Apple's lossless format from iTunes, but

A - If I'm going to pay extra for DRM'd lossless, I better get the cheap lossy version for free (for my phone, wife's iPod, whatever) because paying them to compress a song for me is ridiculous,and
B - It will be a moot point if the player won't play all the FLAC I already have, because I won't own the player. It's why I don't own one now.

Re:24/96?

[Reverberant]

What about 24bits/sample, 96K samples/second?

Enough with the 24/96 wet dreams. Yes, 24/96 does offer real advantages for mixing houses in terms of being able to normalize levels generated by different sources and reducing the complexity of filters. But 16/44.4 is perfectly fine for home audio playback.

What does >16 bits get you? More dynamic range. BFD. 16 bits gets you (realistically) 90+ dB of dynamic range. Unless your listening room has a background noise level of 20 dB or less (trust me, it doesn't), you're not even enjoying the true benefit of the 16-bits you have now.

What does > 44.1kHz sampling give you? Wider frequency response. BFD. Let's assume that most people have good hearing beyond 20 kHz (very few do). Let's assume that most music/movie content has lots of information above 20 kHz (some do, most don't). Let's assume that your speakers can reproduce signals above 20 kHz (some can, most can't). There is still the issue of how you get that > 20kHz info on your recording on the first place. You see, most microphones don't record signals out that high, and of those that do, they only do so over a very narrow angle. When we have tech that can produce mics that are omni-directional above 20 kHz for reasonable costs then maybe you'll have an argument.

Let's deal with the loudness wars before we start worrying about 24/96.

Lossless piracy?

[MMC Monster]

Am I out of the loop? I was under the impression that most piracy was of the low quality mp3s that suck on any high end audio gear.

Lossless is a great idea and may open up a new market to the iTMS, but I can't image it's going to offset piracy. I'd think it will offset physical CD sales.

Re:Or they could just stick with CDs

[michaelmalak]

I do realize they still sell them, but are they $0.99 per song cheap?

I assume you're referring to 45's. Adjusting for inflation, $0.99 in 1979 (the year I bought the 45 of the song "Funkytown" at Woolco) would be, according the the BLS calculator [bls.gov], $2.87 today.

Around 1990, there were CD singles. Granted, they were intended to be replacements for 12" maxi-singles and not 45s, but they were $5. And the record companies killed them because they thought CD singles were "too cheap" -- that they were canibalizing the sales of CD albums.

The expectation of paying $0.99 per song is not based upon historical price trends, but rather upon the expectation that music should be free or at least cheap, which in turn was caused by the record companies withholding digital sales until just a few years ago.

For a studio

[pavon]

That is a good argument for why a studio should sample at a rate that accommodates the roll-off in their analog low-pass filters. However, once that is done you can use a can use a digital lo-pass filter / downsampler which can easily be designed to have very sharp cut-off rates. There is no reason at all for a consumer format to be more than 48kHz.

Re:24/96?

[Brett Buck]

There is an advantage to higher sampling rates, but it has nothing to do with the frequency content of the recorded material or Nyquist's theory . If you sample at 44.1 khz (CD standard) you get 44.1 khz noise in the output. That has to be filtered out somehow, without affecting the in-band audio signal. Rolling off many DB in a short frequency span (factor of ~2) takes quite a filter, which depending on how it's done, introduces phase shifts of the in-band signal. The sound quality from CD players it largely determined by how, and how well, the D-A conversion (which has a frequency response all it's own determined by the guts of the converter)and analog filtering are done.

      Sampling at higher frequencies makes it easier to build a good output filter. That's a very secondary or tertiary level effect, so it doesn't really make much difference, but it theoretically could.

      Note that this is assuming the standard PCM encoding. "Single Bit"/streaming encoding (like SACD runs at fantastically higher sample frequencies, but the frequencies aren't really comparable (and it's not a good way to go because you introduce other issues (like tons of quantization noise).

        The only identified issue with the standard red-book CD format is the dynamic range, but there are so few sources that need more than 16 bits and certainly very few playback systems/environments that will let you take advantage of it, it's essentially a non-issue. HDCD (which is a 20-bit PCM format) addresses this but hasn't and probably won't become common.

    Bottom line - the guys who came up with the audio CD sampling format pretty well knew what they were doing and there aren't any practical limitations in the recording format. Everything else in the system (from microphone to engineering to speaker) is the limiting factor.

        Brett

Re:"Lossless"? Such BS

[Anonymous Coward]

Sorry, Nyquist's theorem states that you can accurately represent frequencies up to 1/2 the sampling rate.

And just how good a representation of a 20 KHz or 21KHz wave does a 44KHz sample capture? Any beating effects? And how much music is a pure sine wave as opposed to superpositions of multiple waves that may have partial transient transitions of >22KHz?

As for 16 bit, nobody uses all that dynamic range anyway.

Oh, so Nyquist's theorem doesn't assume an infinite precision sample and that dynamic range isn't a factor in a more accurate representation of small frequency variations? I haven't looked in detail at the math proofs for Nyquist's theorem but your interpretation doesn't match what I remember about SHM and wave superposition.

Well, one thing's true enough, very few popular recordings use the high end of the human hearing range now that Mariah and Witney are no longer in vogue. With an increasing amount of hearing loss in the general public, fewer customers of popular/contemporary music can hear much above 12KHz, at which point CD's are reasonably satisfactory.

Personally, I'm beginning to wonder if the real reason for dynamic range compression is so that customers aren't surprised by how crappy some manufactured idol bands and singers sound in person without heavy studio voice processing.

Re:sampling frequency and signal frequency !mix-up

[Grishnakh]

That said... the sampling frequency shouldn't be mixed with the signal frequency in the way you mention; e.g. 44.1KHz, divide by 2 (yay Nyquist), ~22KHz is the maximum frequency you can sample. ergo: 96KHz allows you to sample 48KHz signals and nobody can hear 48KHz anyway so what's the point.
Ah, true, but...
A 400Hz sine wave is now -also- sampled at the 96KHz level. Suddenly, that sine wave is looking twice as smooth.

No, it's not. If a 400 Hz sine wave is sampled at a mere 800 Hz, it can be reproduced perfectly, as long as your equipment is designed properly. Again, go read about Nyquist's Theorem before spouting falsehoods.

If you're reproducing that sine wave in a "blocky" way, your reproduction equipment is faulty. There's nothing wrong with the recording technique, because, again by Nyquist's Theorem, no information has been lost.

Re:"Lossless"? Such BS

[jabuzz]

Except you are presuming that the human ear perceives a 20kHz sine wave, and a 20kHz sine wave plus a whole series of harmonics identically.

The problem with that notion is that the standard test for hearing perception is to play pure sine waves of varying frequencies and ask the listener if they can hear them. However over the millions of years of human evolution, it was not until the invention of the tuning fork in 1711 that any human ear had heared a pure sine wave. Up until that point it had evolved to distinguish multiple frequencies at once.

I am not aware of any scientific studies into whether the human ear is able to perceive the existence of harmonics in sound waves above what is considered the normal hearing limit. Surprising really because if it is, it would explain a lot when it comes to sound.

Re:"Lossless"? Such BS

[marcello_dl]

> Those higher harmonics cannot be accurately represented with a 44.1 khz sampling rate, but since you can't hear anything above 22khz anyway it doesn't matter.

"The ear can't pick fundamental sounds at more than around 20 khz" != "the ear does a fourier transform and discard all harmonics above 22khz." The signal processing that a ear does to localize and identify sounds is a little more sophisticated.

I didn't do a double blind test, but even a seemingly small difference between a DAT recording at 44.1 and at 48 khz seems to make a slight difference in the sweetness of high end.

Re:Flac is gay..

[jacksonj04]

It's still a lossy format which strips out some of the audio detail. I'm no gold-connector-magnetically-balanced-shielded-cable audiophile, but I do appreciate being able to listen to the entire depth of a piece of music (especially classical).

Perhaps a better way of putting it would be 'the human ear cannot distinguish between 320kbps MP3 and FLAC if listened to on iPod headphones', which is fair enough. There's no need to include everything if all I'm going to do is listen to it on the bus. Which leads to my original point - MP3 is lossy. AAC which is my format of choice is better quality for the space and bitrate, but is still lossy. FLAC isn't, which means I could have my lossless FLAC copy on my desktop where there's easy storage space, then have iTunes automatically create reduced quality versions for carrying around on my iPod. Compression from lossless source is always better than compression from an already compressed copy.

Not to mention that the iTunes store *isn't* 320kbps. 128kbps for the normal content, 256kbps for iTunes Plus.

Mostly agree, but...

[Gage With Union]

Definitely agree on the sampling rate. For all those golden ears who insist on the higher sampling rates, they should also know that side by side double-blind listening experiments conducted have established that the higher sampling rates sound better not because of the increase in data, but because of the smoothness of the waveform. They pitted 48 kHz recordings upsampled with interpolation to 96 kHz with 96 kHz recordings of the same material and found that there was no discernible difference. This was recounted to me by a Tannoy rep. The real advantage of 96kHz is in the signal processing domain, particularly for plugins that use filters or delay. (having to do with interpolation)

I definitely disagree about the dynamic rate in certain cases. (I'm an audio engineer, DSP programmer, and electro-acoustic musician) The problem with dynamic range is that you have the same number of bits to represent 0 to -6 Db (mono) that you do for -6 to -inf Db. Once you get down to the softest sounds, you often don't have many bits left at all to represent the sound; you only get the full range for the loudest of sounds, and the distribution of bits is linear, while the distribution of loudness is logarithmic. For most music, it's not a problem, but it does cause problems for orchestral works which can have a huge range of dynamics. For instance, in Messaien's Éclairs sur l'au-delà there's a bit with two triangle players on opposite sides of the stage playing rolls as soft as possible, and that never sounds right in recording. Also, there's plenty of electroacoustic pieces that benefit from the increased dynamic range.

If you record at 16 bit but allow 12 Db of headroom just in case of really nasty spikes (which can definitely happen with an orchestra), then you are now effectively recording with 16384 possible values instead of 65536, meaning only 84 Db of resolution with which to record. For most listeners, though 16bit/44.1khz is fine, and it absolutely destroys vinyl in terms of fidelity, aliasing, etc. be damned.

Re:For a studio

[pavon]

Like bentcd said, that is a completely adequate waveform. If you output that using a perfect zero-order hold and a perfect anti-imaging filter, you will get a perfect 24kHz sine wave. Of course, you can't build a perfect hardware, and having more points than the Nyquist limit will relax the requirements on your D/A circuit. However, if you filtered and sampled the data correctly to begin with (ie no aliasing occurred), then upsampling will give you a waveform that is identical to what you would get by oversampling.