Hi
So I’m slightly considering to purchase SL with the 60% off voucher, but I’m not sure if it’s worth it for me.
The problem that I’m often facing is having a particular audio file with some dialogue, where there is some unwanted music or ambience in the background.
Is this easy to fix using SL, while still preserving a somewhat decent voice quality?
Yes, IMO it is the best tool available for this type of job.
But while it gives you the best results, it is not completely automated like other tools.
There is almost always some manual work involved, but the upside is that you don’t have to live with whatever artifacts the process generates, like you do with dxRevive, Dialogue Extract, etc.
You can run Unmix Noisy Speech on your material and then make a manual spot fixing pass for unparalleled results.
There is a free trial version. I recommend you test it first with your audio files.
So I have been using SpectraLayers for this very purpose a lot and you will probably still get some artefacting from the unmix process (all tools will) but with SL you can easily, quickly and effectively process the vocal layer to sort this out. This for me is the key difference, not only the ability to separate the vocals but then easily clean up artefacting from the process, it can even add missing information back with varying degrees of success but in the main with am amazing results.
@reticulatedpython How exactly are you removing those artifacts (or artefacting, not sure how this is spelled)? Which tools (or parameters) do you use for that?
My experience is that if it’s a recording of someone speaking clearly and it has subtle low volume ambience, then doing unmix soundtrack works almost perfectly.
But the moment the background music (or sound effects) is more prominent then the result of separating the elements (sounds) becomes unpredictable, which is understandable, but I’d still like to know if there is a way of polishing the final result and making separate sounds sound a bit more natural.
Usually what is done is that you then go into the noise layer and manually find and and move all the bits and pieces of dialog that mistakenly went there. Move them to a new layer so you have more freedom to manipulate those bits should you move unwanted stuff along with the good ones.
For this task a few selections and tools that come to mind are Transfer Tool, Frequency/harmonics Selection, Wand Selection, Lasso Selection and Cut Special (to new layer/last target).
You can use “sample all layers” so you can count on the Speech Layer to help keep your selections from wandering into unwanted stuff.
Frequency Repair is great for when the noise has completely masked the speech and you need to patch it up.
What pollutes the results are not exactly “artifacts” cause they are not new things that the process creates, but rather pieces of information that get to where you don’t want them to be.
EDIT: Oh, I talked about how to get stuff back, but not how to further clean the Speech Layer. For that, Reconstruct (former Heal), Attenuate and the good old Eraser (which you can think of as an ultra-precise EQ, both in Frequency and Time domains) are my usual suspects.
@henrique_staino has summed it all up pretty concisely, there are some very good tutorials on YouTube, one guy if I remember does voice work and has clear instructions for everything I think you are after.
Hope that helps.
I have a voice recording with some subtle noise from the microphone that sounds like coil whine. What is a good method for removing the noise?
I’d probably run Unmix Noisy Speech first, select the harmonics in the noise layer and run Reconstruct on the selection. Depending on the characteristics of the information present in the noise layer different selections may work best: Harmonics Selecion, Frequency Range Selection on each harmonic, or even just select the whole span where the harmonics lie.
I worked on a short film recently, one of the scenes was in a flea market, and one of the stands were playing a very cheesy dance music, very loud. Also it was quite crowded. And the scene had a lot of movement so there were a lot of doppler and dynamic reverb in that music. I used unmix soundtrack module on those first, it did a terrific job, seperated most of the music, and then with broadband noise reduction (with unmix noisy speech) the result was acceptable. The only issue was, after the unmix soundtrack process, the music left resonances on overlapping frequencies in dialogue, those needed manual cleanup. The director was super happy with the result.
Which tools or method did you specifically use for the manual cleanup? I just purchased the software, so I’m trying to learn the basics.
So far I tried using modules like Unmix Soundtrack to separate voices, sounds and music from each other, and I’m wondering how to deal with the weird sounding voices, sounds or music, that got affected after the separation process.
I appreciate the detailed reply. It’s just a lot of info at once, right now. But it’s good to have the information for later.
I’m trying not to overwhelm myself now with too many features and just focus on a few important tools.
Awesome, thanks! I will check out those tutorials.
Well there is no single answer on this one. My memory about this is, most of the artifacts from music were on overlapping frequencies with dialogue, on dialogue layer, they didn’t sound like music at all, but resonance. Earpiercing bits of audio. Reconstruct module, also simple gain down on some solved most of it.
Found myself back in SLP today after try to mix sections…my pre-plans for surround separation require SLP re-edits to get certain sounds in the right places…
@Aleque
to work manually in the Spectrogram, one really needs to be able to read it to some degree. How are you at knowing what you are looking at? If it’s just all fuzzy TV snow for you, the learning curve is steep and the only way I know how to learn to read the transforms in the spectrogram is study and practice. It has taken me years to get as good at it as I am right now. Used to take me weeks, now it’s hours or days.
I have written A LOT posts in this forum on manual work in SLP.
Number one: create duplicates and edit the dupes and mute your original inputs/separates.
I mostly use selection brush in “Add to selection” mode. You have to know what mode your selection tool is in: default is Replace. You need Subtract too; I never use Intersect. I also use marque (rectangular/ elliptical), magic wand and time. I never use frequency selection tools anymore, I find freehand selection gives much better control.
You need a plan for layers and what goes where. I always work the same:
_I never use erase; I have a trash layer which is muted (yellow)
_voice layer is always magenta
_wanted noise is cyan
I don’t do much music separation and music can be a challenge.
One over all thing I have noticed is you don’t need to be as precise as you might initially think; rougher, wider selections can yield cleaner sound IME. Also, horizontal selections that are static at frequency are less noticeable than a rising or falling (diagonal) selection.
hope that starts you off.
Also, full removal seldom works IME, it sounds like artefacts; noise removal/ redux works great manually. We are still in a situation where we are looking at a rebalancing rather than full stem separation IMO.