#60 Clicks on beginning of a track, when converted to PCM

[Anonymous]

Hello,

I am using latest beta version of Foobar2000 and also latest 0.8.4 version of SACD plugin.

However, when I attempt to convert an SACD ISO - and here it doesn't matter if I load the ISO in the Foobar, or I previously convert via Sonore GUI to DFF/DSF - to FLAC (and/or WAV for instance), I get a very annoying "tick" on the beginning of many tracks.
I am currently unsure if it's fault of SACD plugin or something else.

I haven't got much experience with SACDs, ISOs I got are merely to test the surround capability of my system for music. Is it normal that SACDs have clicks between tracks (or at the beginning of a track)? I do read some posts on the net about this phenomenon... also, the problem existing with older versions of the plugin and DSF files, seems to have been resolved, besides, it's happening with direct ISO-conversion and with DFF.

In the Audacity, the "phenomenon" looks like this:
https://dl.dropboxusercontent.com/u/31158446/Tick-Problem.JPG

And, it's happening with most of the tracks (if not all?). The only program I found out converted the DFF cleanly into FLAC is "AuI ConverteR 48x44". HOWEVER, comparing the files in Audacity has shown that AuI actually shortened the track by a sample or two, basically the length of the tick (see picture). So, I'd call that cheating, not a solution.

Here is a also a pic of my conversion settings in Foobar:
https://dl.dropboxusercontent.com/u/31158446/Foobar-Settings.JPG

The conversion settings, as far as I can see, don't play a role, since I tried with external flac.exe, internal FLAC encoder and WAV. All produced same result.

It seems to me like the DSD -> PCM conversion is flawed.

I posted this on the hydrogenaudio forums, and one answers from the admins was following:

This would be caused by the downsampling process, and the limitations of the storage format itself. DSD is limited to fully positive and fully negative, with no state to describe centered or 0, other than dithering a 50/50 mix of ones and negative ones, which are one and zero bits.

Technically, the downsampler should probably be employing a start state of 0, and employ a slight high pass filter to reduce DC offsets like this.

So any ideas what's going on here? Could this be fixed with Foobar?

Many thanks!
Kosta

[nAr]

Hello,

Wich amplitude in dB has the "tick" you show on the screencaps ?
I confirm I got some similar findings with the PCM decode by the past.

For now, try this workaround :
1) Put the plugin in DSD mode, not PCM. Restart Foobar.
2) Then add a DSP in the chain to downsample to 32/24bit (...) and whatever sample rate you need (192000, 96000, 88200, 44100 ...) - I recommend SoX or SSRC X -
3) Convert to .wav and look for the spike at the waveform at beginning of the tracks ...

Best regards,

nAr

[Anonymous]

I tried another workaround, which was suggested to me on hydrogenaudio forums:
Installed version 0.7.8 and ran the converstion with "fixed point" setting. That didn't cause the said error in the beginning - however I only checked "scientifically" (eg. in Audacity), and not audibly - currently only via RDP. I will also see what your suggestion brings, but I do question now two things:

• why is the "fixed point" conversion no causing such errors
• what is the difference between fixed point and floating point
• what are changes from 0.7.8 to 0.8.4 qualitywise, if any?

[Anonymous]

Look here:
This is a difference when convering DSF in FLAC with the Version 0.7.8 (fixed) vs 0.8.4 (floating).
The anomaly in the lower curve is exactly the problem I am talking about:

https://dl.dropboxusercontent.com/u/31158446/Tick-Problem-Comparison78-84.JPG

Would it be possible to correct this for the v0.8.5? Maybe implement fixed point again?

Many thanks!
Kosta

[nAr]

Hello,

I'm not so good at explaining things. Maxim will do it better than me ! :)
But basically, those filters are low_pass filters and have different schemes to get rid of HF noise.
Different bandwith extensions, but also will exhibit different behaviour in the time domain ...
So between fixed and floating, it might be another thing, all maths and calculation are done using the floating 'comma' instead of fixed point. I wouldn't be able to tell more.

Try the way I said to get at least proof that it works also for you ;)
And let's wait Maxim to explain maybe where the thing is, and if he could fix that in the next releases.

If I look at noise spectral density in the HF, it seems all ways to downsample create some aliasing / might be the 'tick' culprit, or even try to use DXD (32float/352800/384000) in PCM conversion, it should be free from aliasing ...

Best,

nAr

[Anonymous]

Yes, it was my intention to try it. However it was possible only now - and I did it now.

It seems perfect! I checked both audibly and visually (Audacity), and I see no anomalies.

Many thanks for the help!

Now maybe Maxim can fix it so that it works with PCM-mode properly!

[Anonymous]

Sorry, my bad!! I still had 0.7.8 installed while testing with DSD + DSP.
After I installed 0.8.4, same ticks reappeared!!

[nAr]

Some DSD don't produce spike, some do. I don't know but I suspect it's embedded in the DSD tracks. For example, SACD "On Air" iso, A.Parsons "Blue Blue Sky" 1st track, converted to PCM 32/352800, doesn't exhibit any spike. Screencaps below. I'm using 0.8.4-1 version of SACD decoder, with DSD mode, and a DSD SSRC X in 352800 "Otachan Ultra".

Some DSD2PCM conversions do exhibit a spike at beginning of tracks though. Since some DSD produce spikes and some not, I assume this is not a sacd plugin problem.

Best,

nAr

[Anonymous]

But then why does "fixed point" in 0.7.8 never cause a spike, in any of the tracks?? I just converted a bunch of ISOs and I am yet to find the problematic track.

[Maxim V.Anisiutkin]

Could you give me info (album name, track name) where it clicks?

[Andrew]

One example of such album is SACD Rush - Counterparts remastered 2013 (Audio Fidelity AFZ 152). Most noticeable click can be heard between tracks "Double Agent" and "Leave That Thing Alone" when converted to PCM. Click is presented in 0.9.3 floating point, but not in 0.7.8 fixed point.

[Maxim V.Anisiutkin]

0.7.x line used a different approach to fix clicks. 0.9.4 replaces the first and the last sample values with the adjacent values. There are other more advanced methods, but they hardly be implemented in the nearest future.