Hello people,

I am about to record some of my LPs to CD or DVD, which sound better than the corresponding SACD or CD (For example "Pink Floyd - Dark Side Of the Moon").

I will definetly go lossless, because my system is more expensive than 250€ and I don´t listen to headphones.

My big question was: I wanted to know how much more filespace FLAC needs for 16bit/96KHz than for 16bit/44,1KHz and I made following experiement:
I resampled a existing FLAC-file from 16/44,1KHz to 16/96KHz and compared filesizes:

FLAC 16Bit/44,1KHz = 27,1 MB (28.457.494 Bytes)
FLAC 16Bit/96KHz = 42,7 MB (44.789.760 Bytes)

Why is the filesize so much bigger? I always thought that Resampling of CD-Quality-Files does not create new audio-information (because you can´t ABX it) and that actually even the (more-)content of SACDs are just some more zeros at the end.

What are these 15MB of more FLAC? FLAC-overhead for 96KHz or more audio-information (which is of course not ABX-able)?

cheers
L01

EDIT: It looks like, it´s more the problem of FLAC:
MONKEYAUDIO only uses 30,8 MB (32.368.237 Bytes) at 16Bit/96KHz, what is 27% less than with FLAC

In simple terms, 16 bits of storage are required for each of 44,100 samples per second of audio data. Similarly,
16 bits of storage for each of 96,000 samples per second of audio data. This would be the result in simple stereo.

this is clear to me: I was just amazed by the (not that good) performance of FLAC in this example:

FLAC 16Bit/44,1KHz = 27,1 MB (28.457.494 Bytes)
FLAC 16Bit/96KHz = 42,7 MB (44.789.760 Bytes)
+ 36%

APE 16Bit/44,1KHz = 24,7 MB (25.910.621 Bytes)
APE 16Bit/96KHz = 30,8 MB (32.368.237 Bytes)
+ 20%

I thought, that resampling actually only adds some more zeros (which is very good shown in the APE-example) and no more audio-information.

cheers
L01

PS I will still prefere FLAC, because it is more portable for me and faster in playback

Well zero is data too.

If you ad a zero in the midle od a group of ones's you are breaking the pattern

besides i don't thing upsampling just add zeroes.
If it did ther wouldn't be much quality difference in different upsampling methods. and on-the-fly upsampling wouldn't requeire much cpu ressources

I think is this quite surprising.

If you take 16-bit data, and add 8 zeros to make it 24-bit data, then (unless the lossless codec recognises this fact), as it removes redundancy and tries to predict the value of each sample, the correction data required to go from predicted value to actual value needs to be much more accurate. Hence the lossless data rate increases dramatically, even though there is no more real data within the source file.

However, if you take 44.1kHz sampled data, and resample it to a higher sample rate, the prediction (on a sample by sample basis) actually becomes easier. There's more redundancy. There are more samples per second, but they don't really hold any more information. A good lossless codec should be able to take advantage of this, without any special programming.

The fact that this benefit isn't realised suggests that FLAC isn't well tuned for efficiency at 96kHz. It seems that Monkey's is - at least some what - and previous discussions on the Monkey's message board suggest this.


What is the point of taking 44.1kHz sampled data and up-converting it before lossless storage? It is not a fair test for what you intend to do Lokutus01, because (I assume that) you will have frequencies above 22kHz if you record your LPs at 96kHz, whereas CD material does not have anything above 22kHz (up-sampled or not!). Hence the lossless compression ratios could be quite different. The noise floor of LPs won't help either!

Cheers,
David.

Hi David,

I know that for LP it would be actually enough to record in CD-resolution, but actually this was not the target of my test: I wanted to see how good the codecs perform and how good lossless-storage on high-quality recording works, that´s why I just upsampled 44,1-data.

For my LPs I will use the higher resolution, because I would like to make some adjustments to the recorded data and then downsample to CD-Quality, so if I run all the filters, there won´t be that big problems with rounding-errors etc.

Cheers
Andreas

Lokutus01,

I wasn't saying that it would be good enough to record in CD quality when recording LPs - it may or may not, but that wasn't my point.

My point was that you can't test the efficiency of lossless coding at high sample rates using up-sampled CDs. The CDs have no information above 22kHz, but your LP recordings will - so your tests of lossless codec efficiency don't mean anything!

If you're going to convert back to 44.1kHz 16-bit before lossless coding, then I don't understand your test.

Actually, I just don't understand! I was trying to help, but I'm confused now - ignore me!


For absolutely top quality, I'd capture at the best my soundcard would allow (24/96), edit and process at that quality, resample carefully to 44.1kHz and dither to 16-bits. However, I've compared, and it sounds just the same to recording at 16/44.1 and processing at that resolution too. Other factors (LP condition, reasonable turntable, careful use of processing) are much more important. I do sometimes process in 32-bit floating point for convenience (because you can't clip), or when processing good quality CD tracks (because the noise floor matters), but that's all.

Cheers,
David.

hey, no problem, you made your point and honestly: Now, I also understood your point

Thank you in any case: I learned some new stuff

cheers
Andreas

1. resampling adds information in the form of noise which cannot be compressed. but note that though you more than doubled the input rate, the compressed size did not double.

2. the FLAC encoder is not tuned for 16/96. you might get better results experimenting with the options (blocksize, lpc order, rice partitioning).

3. in 16->24 bit case David mentioned, just adding 8 zeroes to each sample, the resulting FLAC file will not be any larger but the resulting APE will. however most resolution conversion algo's don't work like that and will dither and/or add noise.

Josh

Thank you Josh.

Do you think there is still room for improvement of the encoder with 16/44,1KHz, too, or is the limit of the FLAC-format reached with the current-compression-ratio?

In any case: I am very happy with FLAC, because it offers great compression with highest decoding-speed (what is exactly what I need for my music-jukebox).

cheers
Andreas