Low frequency roll-off for speech

This Eq curve seems to be popular and very useful for people doing voice recordings, and it certainly helps to achieve ACX compliance. Perhaps this (or a similar curve) would be a good addition to the shipped EQCurves.xml

<equalizationeffect>
	<curve name="LF rolloff for speech">
		<point f="50.0" d="-120.0"/>
		<point f="60.0" d="-50.0"/>
		<point f="65.0" d="-24.0"/>
		<point f="70.0" d="-12.0"/>
		<point f="80.0" d="-4.0"/>
		<point f="90.0" d="-1.0"/>
		<point f="100.0" d="0.0"/>
	</curve>
</equalizationeffect>

Ideally it should be used with a long filter length, but the filter length is not currently stored in the XML. Fortunately it works quite well with the default filter length, but it performs better with a long filter length, so a second feature request is that Filter Length is added as an optional parameter to the XML definition.

+1
Koz

+1 to both ideas, but “LF” is jargon - to a novice it could look to be to do with “Linear Frequency” or “Length of Filter” that they can see in the EQ dialogue.

Is “Rolloff” more gradual at the transition so that is more correct for this than “Cut”? If so would “Low rolloff for speech” be acceptable?


Gale

Both terms “Rolloff” and “Cut” are used in this context. I have a slight preference for the term “Rolloff” as I think it imparts a sense of a “gradual transition”.

That works for me.
I could also live with “Low cut for speech” though that loses the sense of “gradual transition” which is an important aspect of this setting.

What he said. The tool doesn’t just whack off everything below a certain frequency like High Pass Filter. It’s designed for steep, but gradual transition between the lowest expected valuable voice frequency and the highest expected hum pitch. That’s why there are no adjustments (except for the Length slider).

This is the exact best place for each of those equalization values.

My joke is Steve is going to become world famous not for saving someone’s life at Brighton, but for seven equalization values.

Koz

To dip into consumer-speak, I’ve been calling it Steve’s Custom Rumble Filter.

Koz

OK Steve - what do I do with this one - archive it? Move it to Audio Processing?

Or is it already incorporated into releases?

Peter.

It belongs rather to the audio book production sub forum.

It is not yet in for Audacity 2.1.2.
The next release is supposed to be about upgrading to Wx3 rather than adding new features. As this is not related to that upgrade and is not a bug fix I’m hesitant about introducing it now, though you may want to suggest this on the QA list as a low risk enhancement that has had thorough field testing. Would there be much impact on documentation? I see one image here that would need to be updated: http://manual.audacityteam.org/o/man/equalization.html

The question here is whether to include this as a preset in the main release version of Audacity. As such it is a “feature request”.

And as you know well that has not been adhered to…

My suggestion is just to go ahead and incorporate it.

Peter

I don’t think there is any “risk” but a novice may think the curve is similar to the existing “100 Hz rumble” filter and not know which to use. To what extent are the curves hitting the same problem?

Audacity Manual gives no description of the shipped curves, though hitherto that hardly seems necessary.


Gale

They are similar and for similar purposes.
The main difference is that “LF rolloff for speech” is specifically designed for speech recordings (though it can of course be used with other types of material). The corner frequency is a little lower than the 100 Hz rumble, so less damage to the spoken voice, (particularly for deep voices), and somewhat steeper, so more rejection of unwanted rumble.

Ideally the “LF roll-off for speech” should be used with a long filter length, which currently requires that the user reads the manual, but even at default settings it is likely to perform slightly better for speech recordings. Hopefully we will be able to add “filter length” to the XML curve definition in a future version of Audacity.

Did we decide on a name for this curve? Perhaps “Low rolloff for speech”?
I’d like the name to include the word “speech”, as that presents the strongest clue as to what this curve is designed for (ans I think also answers Gale’s question about deciding which curve to choose).

LF-Rolloff can be used in nearly 100% cases where 100Hz would be used, but 100Hz doesn’t hit the design criteria of LF-rolloff. That and you can make 100Hz and High Pass Filter is also available.

LF-Rolloff has the heavy lifting baked in for voice presentations. It would be nice to be able to stop constantly trotting out “how to add equalizer presets.”

The question does come up, though, how do you bake in the Length setting?

Koz

Currently you can’t.

I’d like to see it as an optional attribute of the “curve” tag in the XML.

Currently we have:

<equalizationeffect>
  <curve name="name of preset">
    <point f="frequency" d="gain (dB)"/>
    ...
  </curve>
  ...
</equalizationeffect>

The above would remain valid, but there would be an optional “length” attribute. For example

<equalizationeffect>
  <curve name="name of preset" length=8191>
    <point f="frequency" d="gain (dB)"/>
    ...
  </curve>
  ...
</equalizationeffect>

The current “Save / Manage Curves” will be replaced by the new “Manage > Presets”.
When saving a preset there could be a checkbox option (in the “save preset” dialog) to include the filter length.

That would be a good name. The only thing I did not like was the “LF” abbreviation we had originally.

Gale

+1

Peter

+1
Koz