• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

PK Error Metric discussion and beta-test

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
This is a good example of how Deltawave can be thrown off a bit. In the difference file you hear a fair amount of the music and some of the pink noise. D-wave has slightly shifted by a small fraction of a sample. When the files actually aren't shifted at all.

If you go to Manual processing and change offset to zero, then you get a 76.87 db RMS null (81.340 a-wtd) and a Pk Metric of -81.9 db. The difference file consists of only hiss. I haven't tried it, but other than the very first or last second of the file I bet you would have more trouble hearing that.
 

mehmethand

Member
Joined
Mar 19, 2021
Messages
10
Likes
4
My workflow is like this:
I'm loading these 2 files into Deltawave as referance and comparison files
I'm pressing match button, Deltawaves analysis the files
I'm checking PK Value.
I m making a blind test between these 2 files using Deltawave comparator.
 
OP
pkane

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
This is a good example of how Deltawave can be thrown off a bit. In the difference file you hear a fair amount of the music and some of the pink noise. D-wave has slightly shifted by a small fraction of a sample. When the files actually aren't shifted at all.

If you go to Manual processing and change offset to zero, then you get a 76.87 db RMS null (81.340 a-wtd) and a Pk Metric of -81.9 db. The difference file consists of only hiss. I haven't tried it, but other than the very first or last second of the file I bet you would have more trouble hearing that.

Right. DW usually needs at least 30 seconds to a minute of a track. 14 seconds doesn't give it enough to work with.
 
OP
pkane

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
My workflow is like this:
I'm loading these 2 files into Deltawave as referance and comparison files
I'm pressing match button, Deltawaves analysis the files
I'm checking PK Value.
I m making a blind test between these 2 files using Deltawave comparator.

That's the right workflow. There's nothing but noise in the difference file, and, as you said, the noise is at -65dB. There is an issue at beginning and the end of the tracks. PK Error metric shows it as well, rising to around -72dBFS at the start and the end. It stays below -84dBFS for the rest of the track. PK Error is computed as an average loudness for a sliding 400ms window. If there is a sharp burst of noise or other difference that falls off by 400ms, these will get averaged out by PK Metric, but may still be audible. Here's what the leading edge error looks like, with the peak well above -50dBFS:

1616209083508.png
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
It is the noise and only noise at the beginning making the most difference. If you chop the first 1.3 seconds from each file then you get -76.86 db for the RMS difference and - 82.8 db for Pk metric. When you play the difference file then you hear noise only with no music.
 

Oristo

Member
Forum Donor
Joined
Mar 24, 2022
Messages
67
Likes
70
Location
South Carolina
Wondering how well PK metric estimates audibility of LP tick reduction,
starting with a track with no obvious ticks. Being my first use of DeltaWave,
its usability was a pleasant surprise. My main challenge was determining
how much to trim from the start of one file (25 seconds) so that DeltaWave could match them.
On my i7-8700K 16 Gb Windows 10 PC, waiting for each recovery for displaying waveforms
to try another Trim Front value took over a minute.
Preliminary results in my environment: -50dB PK metric spikes are not easily audible.
DeltaWave vs Wave Corrector
 
OP
pkane

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
Wondering how well PK metric estimates audibility of LP tick reduction,
starting with a track with no obvious ticks. Being my first use of DeltaWave,
its usability was a pleasant surprise. My main challenge was determining
how much to trim from the start of one file (25 seconds) so that DeltaWave could match them.
On my i7-8700K 16 Gb Windows 10 PC, waiting for each recovery for displaying waveforms
to try another Trim Front value took over a minute.
Preliminary results in my environment: -50dB PK metric spikes are not easily audible.
DeltaWave vs Wave Corrector

The files need to be within a second or two of each other. If they are not, simply click on Show button and look at the Original waveforms plot. You'll see approximately how many seconds you may need to trim from one or the other. Doesn't have to be exact, but should bring the trimmed version to within a couple of seconds of the other file.

To speed things up, you can limit the files to 60 seconds or less while trying to determine how much to trim. In general, 2-3 minutes is plenty to use for matching files. Hi-res files will also take longer, since the number of samples in each file can be 8x or even 16x more than the 44.1k or 48k version. DSD will also take longer to process than PCM.

When you say -50dB PK Metric, is this dBr or dBFS units? dBr expresses audibility of the comparison track relative to the reference signal at that point in time. If signal is at -30dBFS at that moment, PK Metric of -50dBr would indicate noise/distortion at -80dBFS. You'd have to have the volume really turned up to hear a quick error -80dBFS.
 

Oristo

Member
Forum Donor
Joined
Mar 24, 2022
Messages
67
Likes
70
Location
South Carolina
is this dBr or dBFS units?
The graph for -50 peaks was in dBFS; peak dBr is indeed -83, and that plot visibly correlates with tick removals not at all.
I had not yet got far enough into DeltaWave to change defaults other than Trim Front. These samples were 3 minutes, 24 bit 96kHz.
 
Last edited:
OP
pkane

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
The graph for -50 peaks was in dBFS; peak dBr is indeed -83, and that plot visibly correlates with tick removals not at all.
I had not yet got far enough into DeltaWave to change defaults other than Trim Front. These samples were 3 minutes, 24 bit 96kHz.

Can't help you with tick removals :) Trim options can specify how much time to trim from the front and/or back of the file:

1681861988695.png


This will trim 2 second from the front of reference, and 5 seconds from the front of comparison. Both files will then be trimmed from the end by 60 seconds. If the original files were 3 minutes each, the files to be compared will be 118 seconds and 115 seconds, respectively.

On the other hand, if you double-click on the 'End' label, this will allow you to specify the actual number of seconds to take from the file:

1681861956573.png


In this case, files are still trimmed 2 second and 5 seconds from the front, but 60 seconds are then used and matched from what's remaining in the file. Snippets from both files will be 60 seconds in length.
 

Oristo

Member
Forum Donor
Joined
Mar 24, 2022
Messages
67
Likes
70
Location
South Carolina
No, I am totally satisfied with DeltaWave tick removal analyses, now that I understand PK dBr better correlates with audibility.
Take + Show usefully speeds getting Trim Front within a second; thanks for the tip.
Still wonder why DeltaWave considers Joan Baez singing simple/periodic waveform, though..
 
OP
pkane

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
No, I am totally satisfied with DeltaWave tick removal analyses, now that I understand PK dBr better correlates with audibility.
Take + Show usefully speeds getting Trim Front within a second; thanks for the tip.
Still wonder why DeltaWave considers Joan Baez singing simple/periodic waveform, though..
Ah, the problem is that there are more than one points where large portions of the two files can align really well. DeltaWave can't tell where the proper alignment point is for such waveforms, and so guesses that it's a periodic waveform (one that has at least a few repeating sections, such as a sinewave, for example). DW has a different algorithm for dealing with such files, so it offers you to do so. You don't have to agree to doing this, but you will be warned.
 
Top Bottom