Apple/EMI Press Conference Coverage

[localoid]

I'm not at all implying such sounds are IDENTICAL to a real flute in a real acoustic space but the general principle of reconstructing complex sounds from simpler constituents.

As it turns out that was a bad example. For once I got my synths mixed up... and I've removed it regardless.

I'm pretty sure my own hearing is just fine... aside from the fact that fluorescent lights, cathode ray tubes and solid state electronics with 60/120Hz AC cycles occasionally drive me batsh-t insane in a quiet room with no other ambient noise to mask them.

So, can you whip-up that accurate "model" of a flute? It should be easy, since all the makings of that instrument aren't "a secret" or the stuff of "magic" -- they're merely simple mathematical algorithms available in textbook form, most all of them contained in a one-page Wikipedia article.

 

[Westside guy]

I don't know if it says specifically. But I expect the files will still be AAC, just at a higher bit rate and omitting the Fair Play DRM portion.

The announcement explicitly says the DRM-free songs will be AAC, at double the current bit rate.

Since AAC is not proprietary to Apple, this means you could take them to any other player that'll handle AAC. The Zune, for one. You could dump iTunes at some later date and start using WinAmp or XMMS on your computer - they'll both handle AAC that's not DRMed.

 

[localoid]

I was kinda wondering how this discussion about digital audio encoding had somehow drifted over to analog matters...

Why bring this up? Because if we have a 100% accurate "model" of how sound and hearing works (as some here claim), then it should be quite simple to use this (complete and accurate) knowledge-base to produce a 100% realistic "model" of any instrument.

 

[MacinDoc]

Does this mean new iPods soon?

I think Steve's comment in the Q&A that "now is a good time" to increase encoding quality because of lower storage costs indicates that new iPods are just around the corner. Of course, that would mean that at least one more generation of HD-based iPods is likely, contrary to some recent rumors.

 

[lazyrighteye]

Really? Then explain to me why ADPCM streams require much less data and can reproduce the same dynamic range as straight PCM?

Not sure? Here's why...

One way to record the amplitude value at a given interval is to record it as absolute, the total value at that given quantization interval... Another way to record the amplitude value at a given interval is to actually record only the difference between the present value and the value in the preceding interval.

Fewer bits of data to reconstruct the exact same information... Simple math.

Hehe...

I guess I'll have to defer to you on this one as I read your reply like 3 times and am still having trouble with the 'math.'

 

[Avatar74]

I was kinda wondering how this discussion about digital audio encoding had somehow drifted over to analog matters...

Apparently, there are quite a few synth geeks among us! Many people nowadays would be like, "Mellotron? Who's that-- a new mash-up DJ?!?"

Or how about those Farfisa triplets from Hoboken!

 

[nemaslov]

Where is this information given? I guess I don't understand that policy. Why charge more for single tracks but not the album?

Since every business decision is made based on $, I'm assuming they will make enough money from selling singles to justify letting albums go for same price.

Either that or they are trying to be ethical, trying to support the album format so artist don't get discouraged about selling only one or two songs (but i doubt that's the reason).

When I was a kid, 45s (singles) were 79 cents to 99 cents and an album was 299 to 4.99 . You always got a deal by purchasing the whole album. Granted you got two songs but ususally the B side was just that until the Beatles released the very first double A sided singe: Strawberry Fields/ Penny Lane. First time radio played both.

 

[localoid]

...Because of the multitude of tapes and tape playback systems inside, the unit was extremely heavy and prone to breakdown. However, it was used by the Moody Blues and others to provide faithful re-creations of flutes, string sections, and choir sounds in the studio and (occasionally) on the road.

"Faithful" re-creations...???

It sounded a little more "faithful" than say a $50 Casio keyboard from a pawn shop, but it sure didn't fool anyone familiar with what a real orchestra sounds like.

 

[carbonmotion]

Wow, this really could be true after all. Man, Steve and Apple are relentless in their pursuit of consumer product excellence. So much for that econ theory that says a monolopgy does NOT innovate. Apple has 70-some marketshare and still doing this!!!! Simply amazing!!

itunes/ipod is not an monopoly... lol

 

[dXTC]

"Faithful" re-creations...???

It sounded a little more "faithful" than say a $50 Casio keyboard from a pawn shop, but it sure didn't fool anyone familiar with what a real orchestra sounds like.

Point taken. "Faithful" was a very relative term back then.

And whaddaya know: I bought my first keyboard for $50, and it was a Casio SK-1.

 

[Avatar74]

Hehe...

I guess I'll have to defer to you on this one as I read your reply like 3 times and am still having trouble with the 'math.'

Ok...

Say you're a digital system encoding a bitstream...

If you have an amplitude value of 14,280 in quantization interval A, in binary this is represented as:

11011111001000

If your value at the next quantization interval is an amplitude value of 14,283, this is represented as:

11011111001011

In both intervals, 14-bit words are required which comprises a total of 28 bits of data.

Now, imagine you're an ADPCM (Adaptive Delta PCM) encoder...

Instead of encoding the two 14-bit values above, the value recorded at quantization interval B = B - A. Therefore, 14283 - 14280 = a change (or "delta") value of 3.

So in interval B your value of 3 is represented in binary as:

11

Since the system calculates the amplitude value of every quantization interval as a difference from the preceding value, the exact same waveform can be reconstructed with much less data. Only the first quantization interval needs to be represented in absolute form, every successive interval's value can be calculated by adding or subtracting the difference from the preceding value.

Now, another thing about ADPCM is that the quantization intervals are throttled such that if an interval needs only two bits that's what it's going to use. In Linear PCM, 14 bits are still used where in this last example of a value of 3, recorded as 11, would consist of, first, ten zeros ending with two 1's.

So in this one interval you've just conserved 12 bits of data to quantify the exact same amplitude. Granted, the ADPCM decoder needs to have the basic math to understand what its reading, but that's what coding schemas do...

ADPCM is a lossless coding schema because of a couple things:

1. The sound itself is not filtered in any way to reduce or cap the bitstream requirements.

2. More importantly, the encoding used is capable of reproducing the original analogue waveform in a manner that is absolutely indistinguishable from the original.

Granted, AAC is not technically a lossless format but it uses the same principles of more efficient coding to reduce the required bandwidth to reconstruct the same analogue waveform. At certain bitstream thresholds, perceptual transparency relative to PCM is entirely possible.

But it is worth noting that the Audio Engineering Society has stated that 128Kbps AAC is perceptually indistinguishable from 16-bit Linear PCM.

 

[sishaw]

Ok...

Say you're a digital system encoding a bitstream...

If you have an amplitude value of 14,280 in quantization interval A, in binary this is represented as:

11011111001000

If your value at the next quantization interval is an amplitude value of 14,283, this is represented as:

11011111001011

In both intervals, 14-bit words are required which comprises a total of 28 bits of data.

Now, imagine you're an ADPCM (Adaptive Delta PCM) encoder...

Instead of encoding the two 14-bit values above, the value recorded at quantization interval B = B - A. Therefore, 14283 - 14280 = a change (or "delta") value of 3.

So in interval B your value of 3 is represented in binary as:

11

Since the system calculates the amplitude value of every quantization interval as a difference from the preceding value, the exact same waveform can be reconstructed with much less data. Only the first quantization interval needs to be represented in absolute form, every successive interval's value can be calculated by adding or subtracting the difference from the preceding value.

Now, another thing about ADPCM is that the quantization intervals are throttled such that if an interval needs only two bits that's what it's going to use. In Linear PCM, 14 bits are still used where in this last example of a value of 3, recorded as 11, would consist of, first, ten zeros ending with two 1's.

So in this one interval you've just conserved 12 bits of data to quantify the exact same amplitude. Granted, the ADPCM decoder needs to have the basic math to understand what its reading, but that's what coding schemas do...

Interesting. Are these interval values perfectly accurate or approximations? If the latter, does error creep in with successive additions and subtractions? Just curious. This level of understanding of compression is all new to me.

 

[Hattig]

I'm sorry but I'm not getting this. Why $1.99 and not $2.50 or $.99? Is this based on some sort of calculation? And what kind of calculation?

I guess he thought "$2 seems a fair price to buy a *single* song, that big hit of the moment, you know the rest of the album will be crap, etc".

And he'd be right. People buy singles with a song and a crappy remix on them for £1.99 over here, which is about $3 adjusted for tax, etc.

He then made the -$0.01 adjustment because it's a retail item.

Remember that an iTunes song should cost $1.09 or more adjusted for inflation since 2003 as well (although I don't personally know the US inflation rates). So DRM songs are $0.10 cheaper than then, and non-DRM high-bitrate are only $0.20 more! (by 2011 the inflation adjusted price would be $1.29 assuming a 3% inflation rate, I'm betting that the price of tracks on iTunes won't change to compensate though, so if you're really cut up about the price rise then wait until then and you won't be paying any less than in 2003).\

What I'd like to see on iTunes is a "Buy any 10/20/50/100 non-DRM tracks for $10/$20/$50/$100". Not albums, just random songs.

 

[localoid]

Ok...
But it is worth noting that the Audio Engineering Society has stated that 128Kbps AAC is perceptually indistinguishable from 16-bit Linear PCM.

And Crest says 8 out of 10 dentists recommend Crest toothpaste. Wonder if those 8 dentists own stock in Crest, or get free samples of Crest, or have some other vested interest (in saying so)... ???

 

[nemaslov]

To me, I think songs are woth $1.99 each. I really would pay that and I'd think that's fair. For the past 2 years I've been getting them on iTunes for $1.00 less than my ideal price. Whoo, good for me!And now I can get them for $0.70 less than my ideal price.

I'm sorry you're unhappy, but I'm going to be buying a lot of music that STILL costs 35% less than what I think it's worth even AFTER this fantastic change.

Although if being unhappy is what you want, why don't you go around talking about how the songs should all cosst 50 cents? That would allow you to be even MORE indignant and annoyed.

This IS CHEAP. Crap I am paying almost four bucks now for a gallon of gas in San Francisco.

 

[nemaslov]

Funny thing is, now I may buy less from iTunes because I can't decide which darn version to buy.

You can BUY the cheaper limited version first. IF you LOVE the song you can upgrade for just 30 cents...

 

[Avatar74]

Interesting. Are these interval values perfectly accurate or approximations? If the latter, does error creep in with successive additions and subtractions? Just curious. This level of understanding of compression is all new to me.

Quantization error can occur in any digital system... CD Audio is prone to quantization error.

If an amplitude change between two given quantization intervals falls below the least significant bit... then quantization error occurs. At lower bitrates, this may sound like false pumping, but it resolved by dithering which adds slight noise to the signal (imagine a high frequency oscillation like white noise but of very narrow amplitude) that effectively randomizes the rounding up or down of the amplitude values from one sample to the next. The noise is less obvious to the ear than the amplitude pumping that would result without it.

Since changes in amplitude produce significant energy differences to which the ear is tremendously sensitive (our sensitivity from 0 to 120dB is a sensitivity of 1,000,000,000,000 degrees of sound pressure), methods to mitigate quantization error are critical... In this regard, 16-bit CD audio is not sufficient, in my view. 16-bit Linear PCM has a potential 65,536 amplitude values per sample/quantization interval....which translates to a dynamic range of 96dB. 24-bit Linear PCM by comparison has a potential 16,777,216 amplitude values per sample, which translates to a dynamic range of 144dB!

Now, it's important to note that the 1/f noise generated by the A/D-D/A converters in a 24-bit system has the potential to be proportionately greater than that of a 16-bit system, and that does eat away at some of that gain in dynamic range... but all else being equal you can see that even 16-bit PCM is not perfect.

ADPCM is an older system than some transform encoding techniques that achieve tremendous compression with very impressive fidelity (MPEG-4 AAC is one such example) but in principle it's the floating point nature of amplitude values and not sample clocking that's the problem. Sample clocking hasn't really been a problem since highly reliable quartz oscillators have been used for internal reclocking in A/D and D/A converters.

For this reason, digital audio workstations use tremendous amounts of processing resources to mix in 32- and 64-bit floating point before mastering engineers prepare the downmix for a lower-bitrate format like CD, DVD, SACD, etc.

 

[EricNau]

I'm assuming entire albums purchased in the past will still be counted as individual songs for an upgrade (i.e. $.30/song not $.30/album)?

I have spent a total of $57.87 on iTunes music, but since most are albums, I'll have to pay $33.60 to upgrade every song.

So if this is the case, I have to pay over 50% of the original price to upgrade my songs - seems a little high, doesn't it?

 

[G-Prime]

I think Steve's comment in the Q&A that "now is a good time" to increase encoding quality because of lower storage costs indicates that new iPods are just around the corner. Of course, that would mean that at least one more generation of HD-based iPods is likely, contrary to some recent rumors.

I really hope so, I need a new iPod especially if I start using 256kbps encoding music.

 

[Avatar74]

And Crest says 8 out of 10 dentists recommend Crest toothpaste. Wonder if those 8 dentists own stock in Crest, or get free samples of Crest, or have some other vested interest (in saying so)... ???

Or I wonder if this is a dubious analogy since it's a marketing statement published by a manufacturer, the veracity of which is debatable... whereas AES' statement was published by AES which is a standards-establishing organization that scrutinizes all aspects of audio engineering in the same way that technical papers are published by SMPTE today which will decide the future standards for digital theatrical projection.

 

[lazyrighteye]

Ok...

Say you're a digital system encoding a bitstream...

If you have an amplitude value of 14,280 in quantization interval A, in binary this is represented as:

11011111001000

If your value at the next quantization interval is an amplitude value of 14,283, this is represented as:

11011111001011

In both intervals, 14-bit words are required which comprises a total of 28 bits of data.

Now, imagine you're an ADPCM (Adaptive Delta PCM) encoder...

Instead of encoding the two 14-bit values above, the value recorded at quantization interval B = B - A. Therefore, 14283 - 14280 = a change (or "delta") value of 3.

So in interval B your value of 3 is represented in binary as:

11

Since the system calculates the amplitude value of every quantization interval as a difference from the preceding value, the exact same waveform can be reconstructed with much less data. Only the first quantization interval needs to be represented in absolute form, every successive interval's value can be calculated by adding or subtracting the difference from the preceding value.

Now, another thing about ADPCM is that the quantization intervals are throttled such that if an interval needs only two bits that's what it's going to use. In Linear PCM, 14 bits are still used where in this last example of a value of 3, recorded as 11, would consist of, first, ten zeros ending with two 1's.

So in this one interval you've just conserved 12 bits of data to quantify the exact same amplitude. Granted, the ADPCM decoder needs to have the basic math to understand what its reading, but that's what coding schemas do...

ADPCM is a lossless coding schema because of a couple things:

1. The sound itself is not filtered in any way to reduce or cap the bitstream requirements.

2. More importantly, the encoding used is capable of reproducing the original analogue waveform in a manner that is absolutely indistinguishable from the original.

Granted, AAC is not technically a lossless format but it uses the same principles of more efficient coding to reduce the required bandwidth to reconstruct the same analogue waveform. At certain bitstream thresholds, perceptual transparency relative to PCM is entirely possible.

But it is worth noting that the Audio Engineering Society has stated that 128Kbps AAC is perceptually indistinguishable from 16-bit Linear PCM.

Could you please elaborate?





Kidding.
Thanks for breaking it down.

 

[chrisgeleven]

I'm assuming entire albums purchased in the past will still be counted as individual songs for an upgrade (i.e. $.30/song not $.30/album)?

I have spent a total of $57.87 on iTunes music, but since most are albums, I'll have to pay $33.60 to upgrade every song.

So if this is the case, I have to pay over 50% of the original price to upgrade my songs - seems a little high, doesn't it?

Where are you getting that? Assuming you bought 57 songs (since 57.87 is roughly that), 57 * 0.30 = $17.10 to upgrade your iTunes music.

 

[EricNau]

Where are you getting that? Assuming you bought 57 songs (since 57.87 is roughly that), 57 * 0.30 = $17.10 to upgrade your iTunes music.

I've bought mostly entire albums, which cost $10 each, but have (on average) 20 to 25 songs.

 

[Peace]

This is a good deal for all involved.But the more important thing is the other music companies are going to be peer-pressured into following suit..

In the long run this is great.

 

[chrisgeleven]

I've bought mostly entire albums, which cost $10 each, but have (on average) 20 to 25 songs.

Oh, I get what you mean.

I wonder if in these cases there will be a version of "complete my album" so you can get the upgrade at a discounted price.