Dialogue for Documentary

SpectraLayers
1

Hello!

I am currently cleaning location audio for a documentary (really a mockumentary). While I do have some field recorder 32 bit float, most dialogue is straight from a Sony and a Panasonic shooting at 16 bit.

The distributor likes the “terrible” feel of the audio as it fits the story, but I still need to clean it up the best I can for intelligibility. Currently, I am separating noise and dialogue and blending noise back in. This seems to be creating too many artifacts. I am on Spectralayers 11 and primarily using “unmix” modules. I thought it would do better at not leaving as many artifacts if I didn’t erase information. Alas, I’m afraid user error is creating issues.

I am looking for best practices processing. Perhaps the order modules are used. I am definitely not expecting perfection. The plan at this point is to have one dialogue track, and one background noise track. The other issue I am running into is live music playing in the background in a few spots. I need to pull those out as well.

Any recommendations are welcome!

Thank you.

1 Like
2

Hey, learning to read the spectrogram and having a non-destructive workflow takes some teething.

I have explained my process many times here in this forum…frankly, I logged my progress here, so you can dig it up if you can stand it

My basic process is I make a mix to bounce from my DAW and work standalone in SL. I bounce one mic and name it LAV and the other I name RIFLE. Having a naming protocol is key and a folder structure…whatever your process…I’m picky and my filenaming works for me

In SL, after making safety copies and muting them; I run Unmix Noisy Speech on each mic which yields two layers:
Speech & Noise

I rename those layers to match my mic, so, for example
Speech_LAV (always magenta in my jobs)
REALIZE THAT
That unmixed noise layer has wanted speech and noise, so, I name that Noise layer “Good Nz”, so:
Good Nz_LAV (always Cyan in my jobs)

The Speech layer will have unwanted noise too, so same process for both layers

Then I create a new layer for unwanted noises (my syntax is Unz)
so, Unz_LAV (always yellow in my jobs)
Then, mute this layer to transfer the unwanteds to.

To clean out the unwanteds, I mostly use transfer at about 45%. I like the Size and Aspect Ratio settings for controlling the transfer brush area. I also use Marquee, Time Selection and occasionally lasso/ polygon

Working this way is non-destructive. Max undo in SL11 is 64 and you will run out fast.

I have found the most efficient way to set your target layer for transfer is to juggle the layers in the layers panel using the mouse…using the drop-down in the tool settings area can be a pain and it’s easy to make mistakes (get source destination layers incorrect).

So, to do that juggling, try to first move the destination layer to the top of the group and then move your source layer to the top afterward. This is because SL automatically selects the top layer in a group as the destination layer; unless the source layer is the top layer which will send the selection to the second layer in the group. Also, working by moving the source layer up to the top lastly, that layer will be selected.

Realize that the unmixed noise layer has plenty of voice remnants and you gonna need those to eliminate/ greatly reduce your artefacts. I’m getting maybe 1% artefacts in my work. If you try to transfer/ select>copy>paste those voice remnants back to the speech layer, that alone won’t be enough to deal with the artefacts…it might be, but I find just leaving the layers to export mixdown works best for most cases.

That said, I do often transfer remnants back to the speech layer; but typically I only use the selection brush and select the harmonics and shft+x them to the speech layer. I find the harmonic and freq selection tools create more problems than solutions. For example, a chuckle often finds its way to the noise layer…and plenty of other noises like waffling

I will stop now, gotta get to doing this kind of work now!

Also, getting your FFT and brightness settings is key.
I like my FFT at 2048
Max Amp at 0
Min Amp at minus130
Res at 100%
Refinement at x4

took me a while find those setting for me.

The one thing I didn’t mention was clone tool. I use the clone tool after removing unwanteds from the Good Nz layer. And that is “destructive” because there is no way to log those changes beyond creating duplicate Good Nz layer and a LOT of time consuming fooling around for pretty much zero benefit. You could “see” the cloned areas though in the Unwanted noises layer…but all the unwanted noises are there too.

I hope this helps

2 Likes
3

ok, I’ll add as I work

Sometimes the crew chatter is only in the unmixed noise layer,

for that, sometimes I create a layer for crew chatter in each mic group.
I prefer the selection brush for moving voice harmonics . You need to set to additive for the brush tool. This way you can listen to just the selection before cutting that selection to a new layer. Then clone to cover the holes if required.

I know @henrique_staino like to use the Heal function; I seldom use heal :slight_smile:

One “trick” I like to use when selection harmonics with the selection brush is to increase the frequency zoom in the time line instead of changing the brush size…I think it is easier to select the higher (smaller thickness) harmonics that way.

Also, again, re-select is ctrl+shift+D…often, I find when cutting voice remnants from one layer to another, it takes 3 to 5 select>cut to layer below actions…so I a lot of re-selecting might be required

2 Likes
4

another thing I find useful for performance of my old 2018 machine is to set the current tool to Marquee as I work through the file because I get no laggy performance with the marquee tool…If I leave the clone active or a large selection brush as the tool, then SL can get laggy

2 Likes
5

Oh my! This is fantastic. And I will be looking at your previous posts. I’m at a point where the director/producer is fine with nearly anything. And with a May 1 deadline, I can understand. But it’s difficult to “unhear” problems. Lol. I really appreciate the help!

1 Like
6

yikes, it takes me ages with SL…when I started I had a 5 min part take one month!..separating and re-balancing different voices can take absolutely ages

1 Like
7

aaand then there is navigating SL

Personally, I mostly set my start point with the cursor arrow in the timeline ticks (one of my biggest pet peeves with SL is this narrow lane)
Set zoom with mousewheel or the overview lozenge and then move the viewable area with the hand in timeline numbers lane.

2 Likes
8

mmmmm…hard to explain in words

if working on a short area…say 8 seconds
set the start point
and then, with the hand in the timeline numbers
move the display area to leftmost on your screen
this helps when playback shifts the display once the playhead redraws to display to beyond the 8 seconds…took me a little while to get the hang of that

1 Like
9

also, I gotta say, I don’t worry too much about those high harmonics voice remnants in the unmixed noise
if you try to improve your high end with those, it can just seem to increase HF noise…

when getting those high freq consonants (C, S, etc), I prefer to use a large round brush and take the whole thing
If find a less accurate selection area can sometimes really help if you have too many artefacts

for the majority of spoken voice, I’m mostly selecting between 100Hz and 7K;
almost everything below 100hz is a unwanted (I mostly remove thuds and wind);
and except for wanted sibilant sounds, most freqs over 7K are wanted noise/ ambience

unless you have broadband RF or mic blasts…or hands rubbing fabric type hi freq sounds can be the hardest/ most time consuming to reduce; especially when in the crucial 3-7K area

1 Like
10

hmmm, so working today, I realize users can limit the juggling of layers for Transfer tool,
!yet requires a little bit more awareness when working;
if the top layer in a group is destination layer, then highlight source to set as source and a bit less juggling

I find the juggling helps me make less mistakes with destination layer and cut to layer below is readily available as I use shift+X very often

1 Like
11

Amazing posts @ctreitzell !!
For a deep tool like Spectralayers, this is the kind of knowledge that one gains thru personal experience and when shared, we here multiply our improvements.
Thank you very much!

1 Like