Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: HDCD Peak Extension (Read 6004 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

HDCD Peak Extension

Let's put some rubber on the road (in the form of ABX results, of course)...

[attachment=8205:HDCD.flac][attachment=8206:Non-HDCD.flac]The samples are already time-aligned and level-matched*.

(*) these files were cut from tracks that were level-matched using RG.  The files them selves were not level-matched and do not contain RG metadata.  Scanning these files with RG will reveal a 0.11 dB difference, which may or may not affect audibility.

HDCD Peak Extension

Reply #1
Cool.. Just tried this, fairly new to abx testing. Got a mediocre result I guess.
I've been curious about whether something like this is audible as well. The TT dynamic range meter foobar plugin I have shows DR10 and DR14 for the two samples.

Quote
foo_abx 2.0 report
foobar2000 v1.3.5
2015-03-16 13:32:22

File A: HDCD.flac
SHA1: 8cea77d060a1c7b79d2a5839fd190f034adc78c8
File B: Non-HDCD.flac
SHA1: 852c6e946988bb77169018984dc9bf247bceb6ad

Output:
DS : Primary Sound Driver
Crossfading: NO

13:32:22 : Test started.
13:34:00 : 01/01
13:34:09 : 01/02
13:34:20 : 01/03
13:34:30 : 01/04
13:34:35 : 02/05
13:34:50 : 03/06
13:34:57 : 04/07
13:35:06 : 05/08
13:35:20 : 06/09
13:35:45 : 07/10
13:35:52 : 07/11
13:36:12 : 08/12
13:36:36 : 09/13
13:36:44 : 10/14
13:36:57 : 11/15
13:37:08 : 11/16
13:37:08 : Test finished.

----------
Total: 11/16
Probability that you were guessing: 10.5%

-- signature --
da4962cb9c7e217e831a3394dd50bedb251fc56e


Switching between A & B at first, I didn't think I could hear a difference in the drum hits.
In the end, I didn't focus on the peaks, but listened to the the "like you" part, and picked whichever had a "fuller" sounding bass guitar as the non-HDCD file.
Might try again later with more resting between trials.

HDCD Peak Extension

Reply #2
I calculated RG over the entire track and used that value on the sample clips.  RG values for the sample clips differ by 0.11 dB, with the HDCD version being louder.

I'm not saying this difference will necessarily change the results, but people should try ABXing with RG applied.

Sorry about the oversight.

HDCD Peak Extension

Reply #3
Woo, did better on a second try. I checked the box to enable replaygain when starting the test.
Quote
foo_abx 2.0 report
foobar2000 v1.3.5
2015-03-16 15:27:57

File A: HDCD.flac
SHA1: 8cea77d060a1c7b79d2a5839fd190f034adc78c8
Gain adjustment: -2.65 dB
File B: Non-HDCD.flac
SHA1: 852c6e946988bb77169018984dc9bf247bceb6ad
Gain adjustment: -2.54 dB

Output:
DS : Primary Sound Driver
Crossfading: NO

15:27:57 : Test started.
15:30:14 : 01/01
15:31:15 : 02/02
15:31:41 : 03/03
15:32:07 : 04/04
15:33:43 : 04/05
15:34:10 : 05/06
15:34:24 : 06/07
15:34:51 : 07/08
15:35:12 : 08/09
15:35:32 : 09/10
15:36:58 : 10/11
15:37:08 : 11/12
15:39:13 : 12/13
15:39:26 : 12/14
15:39:41 : 13/15
15:40:27 : 14/16
15:40:27 : Test finished.

----------
Total: 14/16
Probability that you were guessing: 0.2%

-- signature --
7e819da4eb6ad6dfa481b27c14e75dee11fe3dcf

This time I was listening to the drum hits on the "dangerous kind" part, and the HDCD one seemed to have more sizzle/abrasiveness.
It was pretty tough, often I couldn't hear any difference and had to pause for a minute.
I also used brighter headphones for this one (Denon D2000, vs HD-600 for the first try).


HDCD Peak Extension

Reply #5
Hmm.. no, I didn't explicitly scan them with ReplayGain beforehand. The only other thing I did in foobar besides importing the tracks and doing the ABX test was use the TT Dynamic Range Meter v1.1.1 component on the tracks.

Track properties for the HDCD file:
https://www.dropbox.com/s/zuvbmd7eta5trgr/S...7%20PM.png?dl=0
(all of the fields in the Metadata tab are blank.)

This is how I had the ABX setup window configured:
https://www.dropbox.com/s/zpxm2gy1ndse1ao/S...2%20PM.png?dl=0

i.e. with the "Use ReplayGain / simple track gain" checkbox checked.
Not sure exactly what that means, but I was assuming it would scan and apply ReplayGain for me. Sorry if I messed up something basic, first time trying to share abx results :-/

HDCD Peak Extension

Reply #6
I was assuming it would scan and apply ReplayGain for me.

You're correct. I didn't know this until I tested just now.

Thanks for your effort!

HDCD Peak Extension

Reply #7
I hear a difference...

Code: [Select]
foo_abx 2.0 report
foobar2000 v1.3.7
2015-03-16 21:19:08

File A: HDCD.flac
SHA1: 8cea77d060a1c7b79d2a5839fd190f034adc78c8
Gain adjustment: -2.65 dB
File B: Non-HDCD.flac
SHA1: 852c6e946988bb77169018984dc9bf247bceb6ad
Gain adjustment: -2.54 dB

Output:
DS : Primary Sound Driver
Crossfading: NO

21:19:08 : Test started.
21:21:47 : 01/01
21:23:26 : 01/02
21:23:45 : 02/03
21:23:53 : 03/04
21:24:02 : 04/05
21:24:22 : 05/06
21:24:49 : 06/07
21:24:57 : 07/08
21:25:08 : 08/09
21:25:16 : 09/10
21:25:24 : 10/11
21:25:45 : 11/12
21:26:23 : 12/13
21:26:42 : 13/14
21:26:49 : 14/15
21:26:55 : 15/16
21:26:55 : Test finished.

 ----------
Total: 15/16
Probability that you were guessing: 0.0%

 -- signature --
97b84d73e0bae580c76a455ff6bee8d7d94603db

... but does this mean I know which one is more dynamic? I believed sample A was a smidge more dynamic, and that turns out to be HDCD.flac in my test... was that the correct answer?

HDCD Peak Extension

Reply #8
I calculated RG over the entire track and used that value on the sample clips.  RG values for the sample clips differ by 0.11 dB, with the HDCD version being louder.

I'm not saying this difference will necessarily change the results, but people should try ABXing with RG applied.

Sorry about the oversight.

As you say, you've already time-aligned them perfectly, which means it's easy to diff them (i.e. subtract one file from the other).

To make them cancel out when diffed (apart from the peaks, which are obviously very different) you need to reduce the non-HDCD version by 0.11dB.

If you do this, then the differences between the two files are all 80dB or more down - except during the peaks themselves, when the differences hit -8dB FS, due to the HDCD version peaking 5dB higher than the non-HDCD version.

I think matching the levels on the non-peaks produces a much fairer test. While the result is that calculated RMS and ReplayGain values become even more different, it means that 95% of the two files are nearly identical, and the only substantial differences are due to the peak extension. This is much better (IMO!) than blindly matching the two files using ReplayGain, which causes the 95% of the file that's supposed to be identical to have a 0.22dB level difference! That's enough to break an ABX test, and must be avoided.

I have uploaded a new version of the non-HDCD file here:
[attachment=8208:Non-HDCD_-0.11dB.flac]
I think people should try to ABX that against greynol's HDCD file. Do not use ReplayGain when ABXing these two files.

If you are interested, you can hear the HDCD file minus the non-HDCD file (i.e. the difference between the two files) here:
[attachment=8210:HDCD_min...non-HDCD.flac]

I didn't use dither when creating these two files. You may prefer to. Or Greynol may prefer to go back to the source and do the level adjustment again in a single step. I can't imagine it will make too much difference.

I can't ABX them.

Cheers,
David.

HDCD Peak Extension

Reply #9
I didn't download the samples yet but what we try to prove? We already had a sample that HDCD with peak extension can sound different when decoded.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

 

HDCD Peak Extension

Reply #11
I tried 2Bdecided's Non-HDCD with no replaygain against the original HDCD.flac and still got a good result, though I'm surprised I did this well, as it was difficult. Again I focused on the drum hits in the second half and listened for a louder *crack*.

Code: [Select]
foo_abx 2.0 report
foobar2000 v1.3.5
2015-03-17 13:34:27

File A: HDCD.flac
SHA1: 8cea77d060a1c7b79d2a5839fd190f034adc78c8
File B: Non-HDCD_-0.11dB.flac
SHA1: 934335400b0217fd57b9203e26232cc1b5fb6cad

Output:
DS : Primary Sound Driver
Crossfading: NO

13:34:27 : Test started.
13:35:53 : 01/01
13:36:07 : 02/02
13:36:49 : 03/03
13:37:00 : 04/04
13:37:53 : 05/05
13:38:32 : 06/06
13:39:13 : 07/07
13:39:54 : 08/08
13:43:05 : 08/09
13:43:17 : 09/10
13:43:53 : 10/11
13:44:23 : 11/12
13:45:17 : 12/13
13:45:28 : 13/14
13:45:37 : 14/15
13:45:45 : 15/16
13:45:45 : Test finished.

----------
Total: 15/16
Probability that you were guessing: 0.0%

-- signature --
7c29a8377f3c62e389696031e1db6102db2541b0

HDCD Peak Extension

Reply #12
Was the level matched properly in either example?

Can we hear it on *this* sample when the level was matched properly?

It was thus drum sample:
HDCD clipping
On that old drum sample we can see that HDCD encoding does not rule out compression. So if music is already compressed by a limiter like in this example the additional cutting of the peaks with the peak extension feature results in more audible distortion.

*this* sample i didn't try yet because i am not at my PC but eric obviously did well already.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!