Diana Krall's strange bitrates

[sjblakey314]

Perhaps someone can help me out with this question...

Lately I've been archiving my DVD-As using Foobar. I've noticed that most 24/48 5.1 tracks come in at about 3500 - 4700 kbits/s when encoded with Flac (level 6). I typically drop the sample rate down to 48, if the original is 96, with SoX to save disk space. However I have noticed one anomaly: Diana Krall's "The Girl in the Other Room" DualDisc. The 5.1 tracks on the DVD-A layer come in at about 1900 - 2000 kbits/s. Finding this odd, I checked back with the original disc - sure enough, the MLP tracks reveal similar bitrates: 1900-2000 kbits/s (or so) for the 24/48 5.1 tracks. However the stereo tracks - at 24/96 are in the 3,000+ kbits range. Does anybody know what this is about? I'd expect the 24/48 5.1 tracks to have similar bitrates to the 24/96 tracks, not substantially less. The 5.1 version of "Temptation", for example, averages 1914 kbits/s for 5.1 (24/48) but 2951 for 2.0 (24/96). What gives?

The only explanation that I can come up with is that someone has reduced the entropy of the 24/48 source, perhaps to something like 16/48, enabling the MLP encoder - or FLAC - to obtain much better bitrates than it otherwise would with a typical 24/48 source. It is strange that this would be done on the 5.1 tracks and not their 2.0 counterparts.

Thoughts?

 

[steelydave]

Do a spectral analysis on the disc, I think that'll give you a pretty definitive answer. I suspect your suspicions will be proved correct.

 

[sjblakey314]

Hi there steelydave, fellow Torontonian,

Thanks for the input. However, wouldn't a spectral analysis only reveal whether there is content above 24 khz? This will work for determining whether a track that is listed as 96khz, for example, has simply been up-sampled from a 48khz. In this case I am suspicious about the bid-depth: I think that a 24/48 track is really a 16/48 track padded with zeroes. I'm not sure that a spectral analysis will help with this.

Perhaps I don't fully understand...

 

[steelydave]

There's an explanation/discussion on the Stereophile forum about how you can calculate effective bit depths from spectral analysis plots. If you do do a spectral graph please post it here, I'd be interested to see it, and I'm sure there are some people here smarter than me that can help you with the math to figure out if it indicates an upsample or not.

 

[LizardKing]

You could try and look at the files in DVDAExplorer it should tell you the bit depth..

 

[sjblakey314]

LizardKing,

Thanks for the tip on DVD-A explorer. I have posed screencaps of its output for track 3, "Temptation" in both 5.1 and stereo, below. Indeed it does indicated that the 5.1 is 24/48 and the 2.0 is 24/96. But look at the compression ratios: 71.2% for the 5.1 track but only 35.7% for the 2.0. Why is the 5.1 track so compressible? My suspicion is that the 5.1 track, while technically 24-bit, doesn't actually have 24-bits worth of precision in each sample.

5.1:


2.0:

 

[sjblakey314]

Well I think I have an answer. After doing some searching I found out about Flac's wasted bits feature. As posted on another forum, this command will reveal the number of wasted bits in a flac file:


flac -ac foo.flac | findstr wasted_bits (for windows)
Source: http://www.hydrogenaudio.org/forums/index.php?showtopic=97746

As it turns out, the results are wasted_bits=8 for the 24/48 5.1 track, and =0 for the 24/96 stereo track. In other words: The 5.1 tracks are actually 16-bit.

 

[sukothai]

It's possible that the 5.1 is more loss-lessly compressible because there is less variation in each channel as compared to the 2 stereo channels. For a simple example, think about the text strings "abcdefg" and "aaabbbb". Both have 7 characters, but the second one is more compressible because it has less variation. The first string can't be compressed to less than 7 characters, but the second string can be compressed to "a3b4" for a reduction in size of 43%. This is a very simple example and with longer strings (or longer audio tracks) you can get even better loss-less compression rates.

I'm not saying this is exactly how the compression is done, but all loss-less compression is done by searching for repeating patterns and less variation always translates to a higher occurrence of repeating patterns. I'm also not saying this is the explanation for this case, but it is a possibility to be considered.

 

[sjblakey314]

I had wondered if this could be explaine by inter-channel redundancy as well. However as most of my other 24/48 5.1 recordings weigh-in at about 3500 - 5000 kbits, it didn't seem likely that this explained the discrepancy. After doing some reading I found that FLAC's wasted bits feature can show if a lower bit depth source has been padded with zeroes. In the case of this DVD wasted.bits=8 for the the 5.1 but 0 for the 2.0. In other words the 5.1 is really 16 bit but the stereo track is true 24-bit
. This seems to explain the etire discrepency.
Evidently this is a not uncommon practice with DVDs and digital downloads.

It's possible that the 5.1 is more loss-lessly compressible because there is less variation in each channel as compared to the 2 stereo channels. For a simple example, think about the text strings "abcdefg" and "aaabbbb". Both have 7 characters, but the second one is more compressible because it has less variation. The first string can't be compressed to less than 7 characters, but the second string can be compressed to "a3b4" for a reduction in size of 43%. This is a very simple example and with longer strings (or longer audio tracks) you can get even better loss-less compression rates.

I'm not saying this is exactly how the compression is done, but all loss-less compression is done by searching for repeating patterns and less variation always translates to a higher occurrence of repeating patterns. I'm also not saying this is the explanation for this case, but it is a possibility to be considered.

 

[steelydave]

Fascinating stuff - great detective work!

I wonder how this happened, if it was intentional or by accident? It's so easy to accidentally export a mix at the wrong bit depth, or perhaps the disc was originally only going to be DVD-V so the mix was exported at 16/48 for Dolby Digital encoding.

 

[sukothai]

Yeah, I have to agree with you. We cross posted; so I didn't see your post about the wasted bits until I had completed my post. When I saw your post I realized you had discovered the real root cause. Thanks for following it through.

I had wondered if this could be explaine by inter-channel redundancy as well. However as most of my other 24/48 5.1 recordings weigh-in at about 3500 - 5000 kbits, it didn't seem likely that this explained the discrepancy. After doing some reading I found that FLAC's wasted bits feature can show if a lower bit depth source has been padded with zeroes. In the case of this DVD wasted.bits=8 for the the 5.1 but 0 for the 2.0. In other words the 5.1 is really 16 bit but the stereo track is true 24-bit
. This seems to explain the etire discrepency.
Evidently this is a not uncommon practice with DVDs and digital downloads.

 

[sjblakey314]

How it happened is a good question. My understand is that the best practice for making lossy files is to feed the decoder (AC-3, DTS, AAC, etc.) with a 20 or 24-bit source, as opposed to dithered 16-bit version; so there is no reason that a 16-bit version would have to be made in the first place. It could be that somebody just made an error. That being said, I found an MLP white paper that alludes to a possible answer. The author suggests that one may want to shave off a few bits in cases where file size is an issue - see s. 9.1: https://www.meridian-audio.com/w_paper/mlp_jap_aes9_1.PDF

Maybe the record company was worried about the extra cost that might result if the file sizes necessitated a dual-layer DVD. There is some discussion on the CA forums about how to scan one's record collection for imposter 24-bit tracks here: http://www.hydrogenaudio.org/forums/index.php?showtopic=97746&st=0

For my part, I've told myself I'm not going to worry about it because I don't have a sound-proofed playback environment with anything like 100 dB of real dynamic range. It is curious, however, that the record company took steps to make the 16-bit tracks look like 24-bit ones to the consumer.

 

[user 4725]

Just out of interest, what is the surround mix like on this album? Are there lots of things happening in the rears or is more ambient?

 

[sjblakey314]

Halloway, this album is mixed as if it were quad. The centre and LFE aren't very busy but there is definitely stuff going on in the rear channels. Perhaps not to the extent of Lyle Lovett's JJR DTS CD (a favourite of mine), where you sometimes have vocal coming at you from the front and back, but there are definately flourishes and interesting details coming from the surrounds, like percussion, tambourine, and even blues guitar - as is the case of "Love me like a man". The surrounds are certainly busier than, for example, Dave Gruisn's "Two for the Road" DTS CD - on which Krall provides some vocals, which I think of as good example of a more "ambient" (though still well done) mix, as you put it.

 

[Cheezmo]

Beck's Guero DVD-Audio is another example of this. The stereo advertised as 24/48 is really 16 bit. I forget if the surround is similarly padded.