Quote:
(all the music files on MusicGiants contain DRM).

Music Giants now offers high resolution DRM FREE files.

RG

Quote:
Elk, can you offer any empirical datapoints on compression rates of 16/44.1 program versus 24/96?

Only the vague approximation that they seem roughly the same, between perhaps 40% to 60% depending on the music.

Quote:

Quote:
Elk, can you offer any empirical datapoints on compression rates of 16/44.1 program versus 24/96?

Only the vague approximation that they seem roughly the same, between perhaps 40% to 60% depending on the music.

Here's some data using the Linn Super Audio Surround Vol 3 sampler which has 16 tracks of various types of music:

CD Quality (44.1kHz 16 bit)

FLAC (Original download) 269MB 46.78%
Wave (Decompressed) 575MB 100.00%
APE (Monkey's Audio) 252MB 43.83%

Studio Master (96kHz 24 bit)

FLAC (Original download) 1.11GB 60.66%
Wave (Decompressed) 1.83GB 100.00%
APE (Monkey's Audio) 1.07GB 58.47%

There would appear to be a significant difference in the compression rates for the high resolution files.

Dave

Geez, I can't find my knees.

I hate math!

Does this mean we are getting some compression on those upper 8 bits which are mostly noise? Say like in the 30% range as opposed to those who claim the noise won't compress at all?

RG

Great examples, Dave. Interesting.

I love the music server idea and will try and hop on board when I upgrade the computer.

Computer audio can be really cool and the i generation has changed alot when it comes to motivation. The sounds cards aren't what they used to be either. Turtle and SB really seem to pay attention to this bracket. Awesome. I believe high end is ready for all that when it comes to computer media including crazy things like the Wadia iDock. Which is kinda cool....

I run some a pair of Quad Active 12L on my computer rig but have done all kinds of combinations of integrated amps and speakers over the years. The computer rig has arguably been the most fun. I like near field listening and headphones so I may be biased.

My FPS games never sounded so good...just kidding, thats silly, but I have actually done some critical listening from time to time in my usual seat as well. The Quad setup, via an older SB Audigy 2 has never failed to make me happy.

I'm sure you could nitpick alot of particulars but I'm a simple man. I never thought my lowly KEF C15 would be moved from their duty but the new rig is really nice for what I've been playin with lately.

Mark

Dave,

Just out of interest are the PCM tracks on this disc stereo or surround sound, i.e. > 2 ch? If the latter then from what I understand FLAC uses inter-channel correlation to increase the degree of compression (presumably by applying the LPC 'across' samples as well as through time although I will need to check the source code to see exactly what is going on). If so the degree of compression here would be greater than for stereo, although presumably the ratio between 16/44.1 and 24/96 files would be similar.

Quote:
Dave,

Just out of interest are the PCM tracks on this disc stereo or surround sound, i.e. > 2 ch? If the latter then from what I understand FLAC uses inter-channel correlation to increase the degree of compression (presumably by applying the LPC 'across' samples as well as through time although I will need to check the source code to see exactly what is going on). If so the degree of compression here would be greater than for stereo, although presumably the ratio between 16/44.1 and 24/96 files would be similar.

Struts,

They are 2 channel tracks; the "Surround" in the title of the collection is misleading.

A 16 bit encoding allows values in the range -65,535 to 65,536 whereas a 24 bit encoding allows values in the range -8,388,607 to 8,388,608. It might be that the more granular representation of the analogue signal in 24 bits yields a data stream that offers less opportunity for lossless compression. More than doubling the sample rate from 44.1kHz to 96kHz will likely add to the impact.

Dave

Quote:
Struts,

They are 2 channel tracks; the "Surround" in the title of the collection is misleading.

A 16 bit encoding allows values in the range -65,535 to 65,536 whereas a 24 bit encoding allows values in the range -8,388,607 to 8,388,608. It might be that the more granular representation of the analogue signal in 24 bits yields a data stream that offers less opportunity for lossless compression. More than doubling the sample rate from 44.1kHz to 96kHz will likely add to the impact.

Dave

Dave,

Based on my high-level understanding of LPC I am pretty sure your second supposition does not hold and possibly not the first either (although I realize I am flying in the face of empirical data here which is never the smartest thing to do!) Will confirm when I have had a chance to study the code.

Dave,

I have spent some long hours poring over the code and I think I have a reasonable idea what is going on. However I seriously underestimated the difficulty of translating a theoretical understanding of the encoding algorithm into practical predictions of how it would perform in a given set of circumstances. The problem is not predicting the algorithm, it is predicting the datasets!

By way of example, in theory the FLAC encoder knows nothing about 'real time', it just sees a stream of consecutive samples representing some complex waveform. Since it has no idea of how many samples correspond to an arbitrary time interval such as a second the sampling rate is effectively transparent and shouldn't affect the compression rate.

Back in the real world things get more complicated. For instance, a 96kHz file can theoretically encode audio frequencies up to 48kHz, however in practice there might not be much signal above, say, 20kHz. In this type of low-pass scenario the magnitude response of the LPC filter tends to deviate significantly from flat and the straightforward per-sample quantization of the LP coefficients that FLAC employs becomes highly inaccurate resulting in inferior compression. (It's all down to known weaknesses in Levinson-Durbin and there is good reason to believe that lattice vector quantization might provide the answer. See this paper for more detail).

So, as I climb this quite steep learning curve I am rapidly realizing that in practice FLAC can indeed do a significantly poorer job of encoding some 24/96 datastreams (it is very data-dependent) than it does with 16/44.1. In fact I have even found examples of cases where FLAC loses it completely and simply passes a given subframe verbatim, which in practice leads to an output (for that subframe) that is actually slightly larger than the input. Go figure!

Thanks, Struts!

Very interesting and makes sense.

Thanks Struts!

I suspect that your understanding of what's going on is way beyond mine and I agree that it is the nature of the source data that will determine the degree of compression achieved in any compression algorithm.

Here is a little more information about the compression rates on the 16 tracks in the Linn Sampler download which might complement your exploration of the code:

- Compressed FLAC file sizes for the 16/44 tracks vary from 31% to 65% of the original
- Compressed FLAC file sizes for the 24/96 tracks vary from 54% to 69% of the original
- The tracks that achieve the highest compression from 16/44 also achieve the highest compression from 24/96 and those that achieve the lowest at 16/44 also achieve the lowest at 24/96 and this is consistent across the tracks with only minor exceptions.
- The tracks that achieve the lowest (worst) compression in this collection are the Jazz tracks (with one exception)
- The track that achieves the best compression is "A Chloris"; a piece with a single piano and trumpet.
- The jazz track that achieves high compression is "A Case Of You"; one singer, bass and piano.
- Compression with Monkey's Audio was completely consistent with the FLAC compression; between 2 and 4 percentage points better on 16/44 and between 2 and 3 percentage points better on 24/96.

Here is a Link to a brief description of the Monkey's Audio compression scheme.

I think the bottom line is that 24/96 files will require between 4 and 6 times more storage space in compressed format than 16/44 files but this will still be around 60%-70% of the space required by the equivalent Wave files.

Dave