Roland de-emphasis

9aae6dc65ef4dff7d302ab29f1c2c2a6

When a signal is recorded by a Roland sampler, a slight boost is applied
to the signal in order to compensate a loss in the high end during the
playback. That’s why it is necessary to de-emphasize the samples before
converting them to other formats otherwise they will sound a little
bit too bright.

The following tests were done with a Roland S-760 OS 2.24, Translator
2.5 build 61 (Windows) and CDXtract 4 (Windows) and with a sample rate
of 44.100 Hertz (The response curve of the Roland de-emphasis filter
changes with the sample rate)

The procedure that I used for these tests is very simple and anybody
who has a S760 can easily verify them.

First I recorded one minute of audio with the S760, then I duplicated
this sample with a new name using the copy function (Sample com / 1:
Copy). I then applied the "-deemphasis" filter to this new
sample (Sample menu / 5:Digital filer) and saved both samples on a Zip
disc. Now you can use CDXtract, Translator or Osmosis to convert these
2 samples to WAVE files (with the deemphasis option turned off). This
way we have 2 versions of the same sample, one with the Roland de-emphasis
and one without. To compare the de-emphasis algorithms you have to convert
again the non de-emphasized sample from the Zip disc but this time with
the de-emphasis turned on. Wavelab has a nice feature in its Analysis
menu called "File compare", it allows you to substract one
sample from another and create a new sample that is the difference between
the 2 samples, this is what I have used in the 4 last screenshots.

The following response curves were exclusively calculated in the digital
domain using filter modelization which gives strictly identical results
than the original filters they copy. The only filter that I was unable
to modelize is the authentic filter from Translator, however I was
still able to make some sample comparisons to check its accuracy.

 
This first graph displays the curve of some de-emphasis filters of Translator.

The red curve is the Roland de-emphasis included in the S-760 (verified
on a couple of units)

The green curve is the standard de-emphasis included in Translator.

The blue curve is the special de-emphasis included in Translator, note
that this is the same curve as the standard one but shifted a little bit
probably because the standard filter was removing too much medium frequencies.
But because the order of this filter is not high enough it doesn’t do
its job at the high frequencies.
This second graph displays the curve of the de-emphasis filter of Osmosis.

The red curve is the Roland de-emphasis included in the S-760

The green curve is the de-emphasis included in Osmosis.
This third graph displays the curve of the de-emphasis filter of CDXTRACT.

The red curve is the Roland de-emphasis included in the S-760 (actually
this curve is hidden behind the green one)

The green curve is the de-emphasis included in CDXTRACT, that’s what we
call an exact de-emphasis.
The following screenshot is the result of subtracting one Roland sample
from the same sample de-emphasized by a S-760. This is the emphasis
part of the signal that should be removed.
The next screenshot is the result of subtracting one Roland sample from
the same sample de-emphasized by the Translator ‘authentic’ filter. So this
is the amount of signal that this ‘authentic’ de-emphasis removes. This screenshot
is at the same scale than the previous screenshot that’s why it’s easy
to see that there is an important difference.
The next screenshot is the result of subtracting one Roland sample de-emphasized
by a S760 from the same sample de-emphasized by the Translator “authentic”
filter. So this is the amount of error of this “authentic” filter. It looks
like the previous screenshot but if you look closely you will see some
differences.
The next screenshot is the result of subtracting one Roland sample de-emphasized
by a S760 from the same sample de-emphasized by CDXTRACT, no comment.