Apply De-emphasis?

[steve]

The only defined points I know about are from the web site at 1061, 3183, and 20000

On the diagram it states "+3dB at 3183Hz, shelving at 10610 Hz, Max at 20kHz.

With the "shelf Eq" in Nyquist, the "frequency" is around the half way point between where the slope begins to deviate from the original level and where it plateaus out. So we are looking for a point between 3183 and 10610.

I set the default gain to -9.6dB but on looking again I think it should be -9.49

The figure of 4000 and the slope of 0.7 produce a curve that look pretty close to that in the graph (when used with a +ve gain).

Not very scientific, but I was just aiming for something in the right ball park
(got to leave you some opportunity to play ).
A bit of manual tweaking should quickly find the best sounding results.

Tweaking Hints:
Increasing the frequency will push the entire curve to the right (x-axis = frequency)

Reducing the slope will make the transition from gain=0 to gain=-9.49 more gradual (flatter curve) so the start of the curve will be at a lower frequency and the end of the curve at a higher frequency.

A high value for the slope (greater than 1) will produce an abrupt jump from the original level (just below the set frequency) to the -9.45dB level (just above the set frequency).

[steve]

Having played around with it a bit more, I think the best values will be around:

gain reduction = 9.49
frequency = 6600Hz
slope = 0.5

but still interested to see what figures you guys come up with.

[aka]

Thanks Steve!

I appreciate you taking the time to offer an assist!
A cursory - things are a bit hectic - listen says even w/ defaults this does a good job - a quick A/B of koz.deemph and steve.deemph files comes up a draw.
Will test and tweak over the weekend. Maybe I'll snip and attach a pre-emphasis'd sample (from Pink Floyd Wish You Were Here 35DP-4 Japan) so interested parties can play along

Humble thanks again, guys!

aka

[kozikowski]

We're both cleverly hiding behind the fact that you need to be three or four dB off before anybody can hear the difference. So any curve that gets close is usually good to go on.

This is the same property of your ear that screws up buying a new living room music amplifier. You have to double the power to hear any significant difference.

That specification of SoX...

<<<The de-emphasis filter is implemented as a biquad; its maximum deviation from the ideal response is only 0.06dB (up to 20kHz).>>>

That gives you engineering bragging rights. It has very little to do with sound or music.

Koz

[steve]

Comments on this:

Code: Select all

<curve name="Redbook De-Emphasis">
<point f="20.000000000000" d="0.000000000000"/>
<point f="50.000000000000" d="-0.001000000000"/>
<point f="100.000000000000" d="-0.004000000000"/>
<point f="500.000000000000" d="-0.096000000000"/>
<point f="600.000000000000" d="-0.138000000000"/>
<point f="700.000000000000" d="-0.186000000000"/>
<point f="800.000000000000" d="-0.241000000000"/>
<point f="900.000000000000" d="-0.303000000000"/>
<point f="1000.000000000000" d="-0.370000000000"/>
<point f="2000.000000000000" d="-1.290000000000"/>
<point f="3000.000000000000" d="-2.430000000000"/>
<point f="4000.000000000000" d="-3.540000000000"/>
<point f="5000.000000000000" d="-4.530000000000"/>
<point f="6000.000000000000" d="-5.380000000000"/>
<point f="7000.000000000000" d="-6.100000000000"/>
<point f="8000.000000000000" d="-6.690000000000"/>
<point f="9000.000000000000" d="-7.190000000000"/>
<point f="10000.000000000000" d="-7.610000000000"/>
<point f="11000.000000000000" d="-7.950000000000"/>
<point f="12000.000000000000" d="-8.250000000000"/>
<point f="13000.000000000000" d="-8.490000000000"/>
<point f="14000.000000000000" d="-8.710000000000"/>
<point f="15000.000000000000" d="-8.890000000000"/>
<point f="16000.000000000000" d="-9.050000000000"/>
<point f="17000.000000000000" d="-9.180000000000"/>
<point f="18000.000000000000" d="-9.300000000000"/>
<point f="19000.000000000000" d="-9.400000000000"/>
<point f="20000.000000000000" d="-9.490000000000"/>
</curve>

[kozikowski]

Is that from the arithmetic? Oh so many microseconds this way and oh so many microseconds that way? Then it's probably far more accurate than the one I did from interpreting the zoomed graphic plus the one or two specified points.

And the poster won't be able to hear any difference in that one, either.

Koz

[steve]

nd the poster won't be able to hear any difference in that one, either.

No, but it looks pretty in the Equalizer window

[kozikowski]

<<<it looks pretty>>>

Oh, yes. I like the way all the "f"s line up and all the least significant data zeros out. You must have been up all night to get that.

Koz

[steve]

Glad you like it but I didn't have to put in all the 0's manually, I got Audacity to do that for me. It was mostly a cut and paste job so it only took about 10 minutes whilst drinking an excellent cup of tea... which reminds me, time to put the kettle on.

[aka]

You are both quite right that the poster cannot hear a difference between any of the 3 (fine) options - but he appreciates the beauty of the final curve, as it were, nonetheless!


Thanks guys!
aka

ps: tried to attach a 30 second sample (flac) but was declined due to size limitation.