DSD2PCM converters produce static noise at the beginning and at the end of the DSDIFF file

[GoogleCodeExporter]

What steps will reproduce the problem?

1. Extracted the DSDIFF files either from the ISO through scarletbook or 
directly from sacd-ripper from non-DST encoded SACD
2. converted the DFF files to PCM using Saracon, proTECH and AudioGate
3. detected the short static noise at the beginning and at the end of the 
converted PCM file both from simple playback and visually using an Wave 
analyzer. 

What is the expected output? What do you see instead?

There should be complete silence (or background tape hiss). The static is 
clearly artificial and is not reproduced during direct playback of the SACD or 
of the DSDIFF file in AudioGate. 

What version of the product are you using? On what operating system?

I used the r191. I also modified the dsdiff.c to exclude all the optional 
chunks, including the ID3 and AudioGate custom chunk, but the short static 
noise is still there.

Please provide any additional information below.

It seems that the problem lies at the end of the header and at the end of the 
DSD chunk (or maybe how the DSD chunk begins) 

Original issue reported on code.google.com by mclaren...@gmail.com on 10 Jun 2011 at 1:39

[GoogleCodeExporter]

Thank you for your report. I've found what the problem is. There is sometimes 
garbage in the beginning and at the end of tracks on certain discs. I'm only 
using the logical sector number at this point and I should be using track time 
within these sectors as well.

So I'm working on it, it will be fixed.. And I will ask you at a later stage to 
retest this. 

Original comment by mr_wic...@hotmail.com on 10 Jun 2011 at 2:02

• Changed state: Started

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

I have to point out that AudioGate produces the short static noise only at the 
beginning of the Wav, and not on all files. Saracon and proTECH, on the other 
hand, act pretty much the same way. 

Original comment by mclaren...@gmail.com on 10 Jun 2011 at 2:12

[GoogleCodeExporter]

issue should have been fixed and fixes have been committed. I want to conduct 
some more testing myself, but if you're adventurous do a compile and check it 
out yourself.

Original comment by mr_wic...@hotmail.com on 11 Jun 2011 at 2:42

[GoogleCodeExporter]

Original comment by mr_wic...@hotmail.com on 12 Jun 2011 at 7:47

• Changed state: Fixed

[GoogleCodeExporter]

Err, I believe this one is not quite gone yet. I still get clicks at the 
beginning of tracks. Not very loud, but clearly audible in tracks with silence 
at the start.

Original comment by lich000k...@yahoo.de on 15 Jun 2011 at 6:24

Attachments:

• sample.dff

[GoogleCodeExporter]

I can confirm what lich000k is stating. The click is clearly still audible both 
at the beginning and at the end of the track. I will try to find "sequitur" 
tracks and check if short-silence is introduced before and after the clicks. If 
that's the case, then it may be a matter of trimming. 

Original comment by mclaren...@gmail.com on 15 Jun 2011 at 8:19

[GoogleCodeExporter]

Can you confirm the "click" is not as audible as the previous clicks?

I've done binary compares of SACD-Ripper output vs. output from Sadie (this has 
limited "extract track" functionality) and the output matches 100%. 

To confirm; are these by accident continuous tracks? Because now I'm thinking 
it depends on the recording as the SACD player does the fade-in and fade-out by 
itself (I know this because I've created DSDIFF masters manually to be burned 
to disc..). 

Although I'm hearing these tiny clicks as well, I don't know what I can do 
other than adding a fade-in and fade-out manually...

Original comment by mr_wic...@hotmail.com on 16 Jun 2011 at 9:05

[GoogleCodeExporter]

I've tested with another SACD with plenty of continuous tracks (Moody Blues - 
EGBDF) and I could not detect these clicks at all. I wonder if it's something 
peculiar of tracks that start and end with silence. If that's the case, I can 
easily live with that, and edit the tracks to get rid of the clicks. Will 
investigate further....

Original comment by mclaren...@gmail.com on 16 Jun 2011 at 10:56

[GoogleCodeExporter]

fyi, here is interesting information regarding fade-in/out on DSDIFF Edit 
Masters:
http://sacd-ripper.googlecode.com/svn/trunk/docs/DSDIFF_1.5_recommended_usage.pd
f

Original comment by mr_wic...@hotmail.com on 16 Jun 2011 at 11:12

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

When I record the PCM output of an modified Oppo SACD/DVD-player (which 
converts DSD to PCM internally) I also have clicks. These clicks seem to be 
linked to the start of each playback, and not to the beginning of specific 
tracks.

Moreover I've read somewhere that digital filters can produce clicking 
artifacts at the start and at the end of the filtered signal (called transient 
response??). This effect depends not only on the filter specs but also on the 
filtered signal, which could explain why they are more or less prominent in 
different tracks.

Therefore these clicks may be an issue of DSD2PCM conversion and have nothing 
to do with DSD extraction. In this case the clicks are kind of a "warm up" 
period for the digital filter used. Can somebody confirm this?

Original comment by thom...@gmx.net on 16 Jun 2011 at 4:56

[GoogleCodeExporter]

> These clicks seem to be linked to the start of each playback, and not to the 
beginning of specific tracks.

This almost sounds as if the clicks could also occur in the middle of a track!
Has anyone experienced this (I haven't)?

But it may also just depend on how long after the beginning of the track the 
signal begins to differ from silence (significantly).

In the latter case the problem might me defeated by setting the marker offset 
field (see PDF from above) accordingly, e.g. 4 frames before the track start.

(Where) does sacd-ripper set this marker?

Original comment by lich000k...@yahoo.de on 16 Jun 2011 at 6:51

[GoogleCodeExporter]

> This almost sounds as if the clicks could also occur in the middle of a track!

No. I suggest that the clicks result from DSD2PCM conversion.
(i.e. the clicks are transient responses from heavy lowpass filtering)

If my theory is true, clicks should occur at the beginning and at the end of 
all DSD2PCM conversion processes, more or less audible depending on the signal 
and/or silence.

As mclaren pointed out the clicks are...

> not reproduced during direct playback of the DSDIFF file in AudioGate.

This also leads to the conclusion that the DSD2PCM conversion produces the 
clicks.

@lich000k:
Can you download some DSDIFF files from
http://www.2l.no/hires/index.html
convert them to PCM and check for clicks?
I don't have Saracon, proTECH or AudioGate

Original comment by thom...@gmx.net on 16 Jun 2011 at 9:27

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

[Edit...]

Ok, I have conducted some tests.
It does indeed seem the clicks are produced in the DSD to PCM conversion. I did 
not have any clicks when I converted the same files in audiogate, but both the 
foobar plugin and Saracon produce clicks.

The foobar output is clearly different from audiogate, as seen in audacity, see 
first picture. This picture shows track 4 from the Babatunde Olatunij ripped 
with sacd-ripper.

However, it does not seem to happen with all files.
The second picture shows a track downloaded from the 2l-site. There is also a 
difference in output, but I can't hear any clicks in either of them.

My guess is the clicks happen in those tracks that have an amplitude much 
greater than zero very close (maybe < 1s) to the start of the track.

The third picture shows a zoom-in on the click.

Original comment by lich000k...@yahoo.de on 19 Jun 2011 at 6:06

Attachments:

• [foobar vs audiogate.jpg](https://storage.googleapis.com/google-code-attachments/sacd-ripper/issue-6/comment-17/foobar vs audiogate.jpg)
• [foobar vs audiogate 2.jpg](https://storage.googleapis.com/google-code-attachments/sacd-ripper/issue-6/comment-17/foobar vs audiogate 2.jpg)
• click.jpg

[GoogleCodeExporter]

It seems that trimming the first 0.0006 seconds (6 tenths of a millisecond) and 
the last 0.0006 seconds of the Wav will get rid of clicks in all scenarios. I 
will do some more empirical tests, but it may be just a matter of few seconds 
using a SoX batch. 

Original comment by mclaren...@gmail.com on 20 Jun 2011 at 10:54

[GoogleCodeExporter]

I'll go ahead and confirm the fixed length, although the 0.00057 sec. trim 
seems to be slightly more accurate. It's very likely at this point that the 
static is used to signal the beginning and the end of the song. I will keep 
testing, but at this point I'm already at the sixth album. These trims are 
performed after the Saracon conversion, which is widely considered the best 
one. 

Original comment by mclaren...@gmail.com on 20 Jun 2011 at 2:08

[GoogleCodeExporter]

Afaik similar clicks can be observed at the beginning of digital SACD rips, 
when the replay is started.

Original comment by gr...@index.hu on 20 Jun 2011 at 2:30

[GoogleCodeExporter]

> It's very likely at this point that the static is used to signal the 
beginning and the end of the song.

But how do you explain some tracks don't show the clicks (e.g. the B.Britten: 
Simple Symphony, Op. 4 - Boisterous Bourrée (2L50SACD_tr1_DSD_stereo) from the 
2L page)?

> [...] Saracon conversion, which is widely considered the best one

Do you have any reference for this, I would be interested?

Original comment by lich000k...@yahoo.de on 20 Jun 2011 at 6:24

[GoogleCodeExporter]

I think we should stay on topic. DSDIFF files converted from DXD source cannot 
be compared with files extracted from SACD discs. I used the Saracon as the 
reference because it is the DSD2PCM converter of choice by HDTracks. There are 
discussions about it, but for the time being let's keep it as "issue #6"

Original comment by mclaren...@gmail.com on 20 Jun 2011 at 7:04

[GoogleCodeExporter]

> DSDIFF files converted from DXD source cannot be compared with files 
extracted from SACD discs.

Maybe the "clicks" are already present in the DSD stream on SACDs, as you say, 
to indicate beginning and end. They would then not be output by SACD players.
But if digital rips from modded SACD players also show these clicks, it means 
the beginning/end information has to be processed by the player at a very late 
stage. Maybe the DAC?

In any event, the best solution would probably be:
SACD-ripper detects beginning/end if available and truncates accordingly.

Original comment by lich000k...@yahoo.de on 20 Jun 2011 at 7:19

[GoogleCodeExporter]

I know that Oppo had a hard time eliminating issues with clicks at the 
beginnings of SACD tracks on the BDP-83.  Maybe unrelated.  Maybe not.

Original comment by chee...@gmail.com on 20 Jun 2011 at 7:43

[GoogleCodeExporter]

I've conducted some tests too. I did not manage to compile the ripper yet, so I 
used a DSDIFF file downloaded from the 2L site.

1. I converted the DSD file to PCM using foobar2000.
2. For comparison I trimmed the first second of the downloaded DSD file and 
converted the rest to PCM.

You can see the results in picture1. Both PCM files show a starting click. 
picture 2 and picture 3 zoom into the clicks.

This clearly shows that the clicks are an result of DSD2PCM conversion! They 
are not present in the DSD stream.

Original comment by thom...@gmx.net on 21 Jun 2011 at 11:52

Attachments:

• pic1.jpg
• pic2.jpg
• pic3.jpg

[GoogleCodeExporter]

Another test... This time I created an audiofile file with 1 sec of digital 
silence, converted it to DSDIFF format using Audiogate to get 1 silent second 
in DSD format.

With Audiogate I merged this silent second with the downloaded DSDIFF file from 
2L and saved the whole as one "silence-extended" DSDIFF file.

Then I converted both, the "silence-extended" and the "original" DSDIFF file, 
to PCM using foobar.

As you can see in the picture, the clicks always occur at the start of the 
conversion process! They are NOT present in the DSD stream. Moreover it doesn't 
make sense to include track markers for beginning/end into an Audiostream. To 
indicate the start/end of tracks you have the TOC, sectors, timeprints, 
subchanneldata...

It is very likely that the clicks are transient responses from lowpass 
filtering during the DSD2PCM conversion process. AFAIK the duration of 
transient responses depend on the filtering order, whereas the amplitude depend 
on the filtered signal.

DSD2PCM conversion is a complex issue, that I haven't fully understood yet. As 
lich000k has elegantly shown, the clicks vary with the DSD2PCM routine used. In 
my testings I could confirm that Audiogate produces less clicks during DSD2PCM 
conversion than foobar. (Maybe Audiogate fades the files after filtering?)

For PCM conversion of DSD tracks, I recommend to combine all DSD tracks and 
convert the whole disc to PCM instead of converting each track separately. With 
this method you just have to trim/fade the begin and the end of the disc and 
not between the tracks. You may then cut the PCM disc image into separate 
tracks.

Original comment by thom...@gmx.net on 22 Jun 2011 at 12:40

Attachments:

• SilenceExtension.jpg

[GoogleCodeExporter]

Off-topic: Out of curiosity, are you aware of an affordable tool which is 
capable of combining multichannel dsdiff tracks without transcoding them back 
and forth?

Original comment by gr...@index.hu on 22 Jun 2011 at 1:10

[GoogleCodeExporter]

did you try audiogate? it is free for download.

Original comment by thom...@gmx.net on 22 Jun 2011 at 1:20

[GoogleCodeExporter]

audiogate does stereo only, sorry.

Original comment by thom...@gmx.net on 22 Jun 2011 at 1:24

[GoogleCodeExporter]

The clicks in DSD to PCM conversion are more or less normal. Silence in DSD 
terms can be a lot of things. In PCM silence is just zeroes. In DSD a so-called 
"silence pattern" is used. This can be the bit pattern 01010101 for example. 
When converted to PCM this gives silence. But also does 00110011 or 10101100 
(which is actually what my Sony SACD player uses when I hit the pause key). In 
short there are several ways to represent silence. 
In DSD to PCM conversion the filters have to be initialized with silence. If 
the converter you use uses a different silence pattern than what is found in 
the beginning of the file you will get a short click when jumping from one type 
of silence to another.

Olaf

Original comment by Olaf.O...@gmail.com on 22 Jun 2011 at 1:41

[GoogleCodeExporter]

Olaf, your explanation seems to be the one that makes the most sense. 
Considering the fact that there seems to be nothing that can be done in 
sacd-ripper that could fix the problem, I'd suggest to close this issue. 
Trimming half-millisecond from the tracks solves the problem in any case. 

Original comment by mclaren...@gmail.com on 22 Jun 2011 at 1:52

[GoogleCodeExporter]

> Moreover it doesn't make sense to include track markers for beginning/end 
into an Audiostream.

Maybe you are right. But maybe by setting the marker accordingly, the converter 
can be tricked into jumping from one type of silence into another (see Olaf's 
explanation, which to me sounds very reasonable), before the track even starts 
and the click will no longer be present in the pcm file.
I have no idea if this is even possible. However, if it works, it is the safest 
and most elegant way to overcom this problem for certain.
Maybe mr_wicked can say something about this.

> You may then cut the PCM disc image into separate tracks.
One of the strengths of sacd-ripper is that it can rip separate tracks. I 
wouldn't want to sacrifice this.

> I'd suggest to close this issue.
Since the clicks happen in the conversion, sacd-ripper is not at fault, I agree.
Still, we might stand a better chance to get the problem fixed in sacd-ripper 
than in saracon.

Of course trimming half-miliseconds avoids the clicks and this process can be 
batched. But what about continous tracks? Are you sure trimming half a 
milisecond is inaudible under any circumstances?
I don't want to manually detect if a track is continous or not and trim 
accordingly, at least not if it can be avoided.

Original comment by lich000k...@yahoo.de on 22 Jun 2011 at 3:28

[GoogleCodeExporter]

I tested the trim with several continuous tracks, and it seems that the trim 
works as supposed. Beside no human can detect a half millisecond skip, and the 
0.00057 comes after some empirical tests as the most "lossless": it just 
removes the short silence and click before the actual track sound. 

Original comment by mclaren...@gmail.com on 22 Jun 2011 at 3:59