Speak UP Speak DOWN | How do you fix that dialogue?

This is not a stupid question or a beginner question but we are all dealing with this and its usually not a big issue but I’m just curious what’s your favorite approach to this:

In a Dialogue, most people have the tendency to speak DOWN, I mean (louder) at the beginning of a sentence and then, slowly start to talk much softer.
Like this:

There are many ways to treat this nasty habit. I’m using (depending on the project) mostly this kind of remedies:

-Dynamics on the track
-Limiting on the buss
-old-school fader (automation) compensation (I use this mostly)
-Envelope processor (depending on the editor)
-cut-and-level (make a cut until the loud part and bring it a bit down)
-“intelligent” processors/plugins

Of course, for every production there is a suitable approach; a basic corporate production gets a faster treatment than a hi-profile cinema/broadcast production.

but what are your favorites? anybody any other insights?

I cut and clipgain. Quite much actually. Maybe even many cuts during one line.

This is kinda a compressor’s reason for existence :slight_smile:

Depends on the project at hand.
If the case is quick delivery, compressor yes.
If something a bit more ambitious, manual editing and mixing.

It’s a combination of all of the above:

  • First I use “Alt” on select tools to cut High or low sections
  • Use “Increment / decrement event volume” key commands (does+/- 1db increments)
  • Use DOP for bad Sections or distortion, etc. (use with key commands for each plug, and presets in DOP)
  • Low cut filter on dialog buss
  • Deesser on dialog buss
  • Use leveler plug in (does no comp) for +/-3 db center @ -24 LUFS
  • Use leveler comp (CL@A type)
  • Multi band Comp like Fabfilter
  • Then regular comp like Pro-C
  • Then Multi limiter like L3-L16

Every comp is subtle, always gain stage @ -24lufs for all plugs… It works wonders ! Super fast editing !

Clip gain adjustments are highly dependent on tone though in my experience. The bigger the difference in speaking volume and higher the noise the harder it gets to just cut shorter sections and adjust level since the room tone gets bumped abruptly at the crossover. So in those cases just riding the fader is easier in my experience.

I typically leave compression to after the manual leveling has been done.

I also find all of this to be highly dependent on the content I’m mixing. Commercials, docs, reality, sports, lifestyle and fiction all have their own requirements, not to mention the various delivery formats. I generally hate listening to harsh compression in a movie theater. It sounds like crap to my ears and is very fatiguing. Obviously something meant for the web and to be played on smartphones and laptops is a different matter…

Incidentally: If there was one plugin function I whish Steinberg would ‘develop’ or ‘add’ it’s maximum reduction level on a compressor. In Pro Tools’ compressor we can set a threshold and ratio as usual but then also select a maximum reduction amount measured in dB. So we can have a knee we like to smooth into compression above the threshold, but regardless of ratio once it reaches the maximum reduction value it will stay at that level. So setting it to 4dB for example means it’d reach 4dB compression according to the knee settings and then even at a 2:1 ratio 12 dB above threshold would output 8dB above, not 6dB… i.e. still 4dB compression, not 6dB…

Nice to see this board is still as constructive as a professional community should be (there are other places where a basic post like this would end up in a b#tchfight)

I’m also happy and mixed disappointed that the tools and workflows we have are still the way to go for most of us, so nothing new to learn here.

Indeed the mix of these options is what everybody does, and compression/dynamics is what every dialogue bus needs to control the stem. The “leveler” as plasmatron opted is a thing I never really dug into but I guess the Waves OneKnob Louder (what I use) is pretty much the same.

There is still one other technique I found myself using sometimes.
I’d like to mention the Envelope/Fade… with a cosine curve at the fade-in you cover the regular fade-in and, when stretched enough, it covers the “speak-up” issue also:

I have most of these things set as a macro and recall them with a EUControl controller, that saves time but we are doing this stuff all the time so I was hoping that someone came up with a more modern hattrick.

I’m also loving the IzotopeRX offline algorithms but wished that kind of stuff would be implemented in Nuendo.

+1 for the good question and informative answers. Thanks, all.

Indeed the mix of these options is what everybody does, and compression/dynamics is what every dialogue bus needs to control the stem. The “leveler” as plasmatron opted is a thing I never really dug into but I guess the Waves OneKnob Louder (what I use) is pretty much the same.

On wave website :Oneknob: Peak limiting and low-level compression with automatic makeupPerfect for individual tracks and full mixes alike

So it does some comp :wink: The leveler is a slow volume checker ! It woks differently !
I use TBProAudio Gainrider 2

Take care !

I think double compression would handle it. Proper settings make the words ends intelligible with no overcompression. Watch some tutorial on youtube, there’re plenty of them and save your workflow learning the way just one time instead of cutting and raising clips volumes forever) Hope it’ll help you.

“Vocal Rider” from Waves should work.

Vocal rider is too random IMO.
Manual clip gain is best I think. It is fast to do and no surprises…

Upwards compression saves the tail end. Otherwise old-school fader riding. Sometimes ride low EQ when you raise too much so it won’t sound bassy.

Soundradix Powair
Thats the best tool for dialog leveling imho.

I’ve yet to find anything better than manual automation + vanilla compression following that. Waverider or whatever they’re called never quite worked for me. I’m not really seeing any time savings if I have to go into spots to tweak. Might as well just give it a pass manually instead.