DACs -- Bit perfect +filters

[radiant_radish]

I did a quick upsample of original/remastered Cocteau Twin tracks and neither clipped. Another apodizing counter villain, Bowie's Blackstar, clipped. I seem to have a large number of recordings that fall foul of this and most of these tracks do exhibit a digital harshness (not just loudness). By contrast, I have the 2016(?) remaster of Bowie's Man Whol Sold The World in hires which is very nice and crankable, apodizing count is zero for that.

 

[John Phillips]

...
You'd have the explain the "apodising counter" comments. Not clear to me what in practice is happening in HQPlayer as I've never used it. Can only *guess* it may refer to intersample overs. But dunno.

A quick search reveals only a little. The author explains:

"... analysis is run on source content to indicate need for apodizing filter. As such it is just an indicator to tell you when you need apodizing filter to correctly reconstruct the source content. If it stays at 0, you don’t necessarily need to use apodizing filter, but there’s no harm at all still using such.​

​

So the sound quality improvement is only through your choice of filter based on the provided information.​

​

It is counter to indicate how many times HQPlayer has detected certain types of distortion inducing errors in the source content. Latest improvements make it more accurate and reliable."​

Like you, I wonder if "distortion inducing errors in the source" means data that reconstructs to above full scale. Would an apodizing filter deal better with that? I haven't ever thought about that issue.

My tests here a couple of years ago failed to persuade me that different filters did anything significantly different. However that was on classical music (and a small amount of jazz) and I suspect largely recorded so that true peak stays below FS.

 

[Jim Audiomisc]

A quick search reveals only a little. The author explains:

"... analysis is run on source content to indicate need for apodizing filter. As such it is just an indicator to tell you when you need apodizing filter to correctly reconstruct the source content. If it stays at 0, you don’t necessarily need to use apodizing filter, but there’s no harm at all still using such.​

​

So the sound quality improvement is only through your choice of filter based on the provided information.​

​

It is counter to indicate how many times HQPlayer has detected certain types of distortion inducing errors in the source content. Latest improvements make it more accurate and reliable."​

Like you, I wonder if "distortion inducing errors in the source" means data that reconstructs to above full scale. Would an apodizing filter deal better with that? I haven't ever thought about that issue.

My tests here a couple of years ago failed to persuade me that different filters did anything significantly different. However that was on classical music (and a small amount of jazz) and I suspect largely recorded so that true peak stays below FS.

What is meant by "need an apodising filter"? I'm not clear what they mean by that.

Do they mean 'checks show that some inter-sample levels are above 0dBFS?

if so, the solution is to either scale down all the samples *before* rendering OR to use a reconstruction filter that can accurately output analogue up to this 'over' level.

Simplest approach: Take all input and shift it down 6dB - i.e. one bit downshift - into values that have enough bits per sample that the lowest bit doesn;t get lost. Then use that series to do the reconstruction. Amp up the output if you wish to compensate.

Then simply do this to all input. No need for any 'detection and blinken lighten' to bother the user or change mode.

Alternative would simply be an output analogue stage with filtering that can cope. In theory fine, but in reality I suspect the above would be easier for getting reliable results at trivial cost.

Not done a reliable survey, but my personal impression is that overs or clipping are more common on pop/rock than on good jazz or classical music. Parallel here, maybe, with care taken to avoid needless level compression rather than use it as a "loudness sells" weapon.

 

[adamdea]

What is meant by "need an apodising filter"? I'm not clear what they mean by that.

Well the starting point would be to establish exactly what an "apodising filter" means here. I've never been entirely clear about this. In some contexts I understand it to refer to the use of a (non-rectangular) window function to remove spectral leakage, but IIRC didn't meridian use it to refer to something which removed pre-ringing ie a minimum phase filter? So the detector detects that there is something wrong which needs to be removed by wrong-thing remover?

 

[adamdea]

Could it be that the detector detects aliasing due to inadequate stop band rejection in the pre adc anti aliasing filter (or decimation filter) and than applies a brick wall filter which reaches stop band at below Nyquist in order to remove same?

 

[radiant_radish]

I'm completely out my depth but I think it may have something to do with decimation - the majority of my hires tracks don't trigger the count. The following SoX recipe was apparently a Meridian style filter:

sox -q -t wav - -t wav - rate -v -a -M -b 90.7

-a Allow aliasing above the pass-band
-M Minimum Phase response
-b band width