So Im working on some old stuff from a client and it has serious bleed on a cajon mic. A cajon isnt really a drum kit so the unmix is not working
If I have a range of isolated hits; what is the best way to train (a dragon?)…would be good to have some decent idea of what Im looking for not just “10 secs”
Should it be a range of hits/dynamics etc to produce the best information…are there tips and tricks that would help here
I think Unmix Song does it fairly well with only Drums ticked.
Some hits and slaps are missing but it’s quite easy to catch them from the ”Other” layer with an Unmix Components on that layer. There you’ll get Tonal, Transient and Noise sublayers which contains useful material.
Most of the Tonal components are from the singer (and a guitar?) but a few seconds in and at the end there are some lower frequencies that belong to the cajon; select and move those to the Drums layer. Keep the Tonal layer temporarily until the later moment for cleansing the Noise layer (below).
Then everything - I believe - on the Transient layer belongs to the cajon slaps, so merge that layer with the Drums layer in toto.
The Noise layer contains some mids and highs that will give the cajon its character so some of this should go back to the Drums layer as well.
One way of cleaning the Noise layer from vocal remnants - mostly low mids - before moving the contents to the Drums layer could be by using an Imprint function that involves both the Tonal and the Noise sublayers, or if that won’t work use De-bleed to subtract the Tonal layer from the Noise layer.
It may seem somewhat involved when you read this but it’s very simple and in my opinion the results are great. I can’t post an audio file, hence the long story…
Thanks for the suggestions…will try but I have tried combinations…chasing my tail.
I just noticed the unmix had important elements missing especially took away the crisp initial finger taps etc
Its made me realise how non musical it is and how much I have veered away from edits and just love the capture and hands on mix
Its kind of my fault as, even though the players were a fair distance apart and its a spot mic on the cajon, the tapa (face plate of cajon) is exacerbating the spill/bleed as they were facing each other hence the bleed has more probs than normal.
As per OP…by this time and having possibility of capturing isolated detailed models, I was hoping this might be a thing
Concerning the training aspect; it’s not possible for us users to train SpectraLayers like in machine learning. The training is done by the software makers.
The Unmix Instrument module is not about training the app to learn instrument models but rather a rudimentary way to guide the app what to separate by means of a short simple tonal selection. I’m not sure how it would work with percussive sounds.
SL was ok before but now it doesnt work the same and makes more work for me.
Most of the unmix requirement is for mostly organic tracks from group/band recordings or drum tracking…not really for artificial stuff (because if its self generated it doesnt need unmixing)…so they are pretty easy targets.
Originally I was unmixing commercial songs to use individual stems (for research and analysis) but it sent me on a bit of a wild goose chase as they need to be pre master for any sort of relevance and in the end, its your ears, context and subjective taste that really matters. Im mixing with may more headroom now and K-14 is about the max for most clients…then just do 2 mixes; 1 audiophile for peeps that want to be immersed and feel it (and have the gear to do that and are able to turn up the volume knob) and the “Slaves to the altar of 0dbfs” who are convinced that physics can be cheated…so beware of the distractions
Just need the tools to help with achieving those outcomes…LALAL has worked so far but thanks for the heads up @DosWasBest
Its def different to LALAL I notice the sizzle of the snares is a bit subdued on your version and when it moves to the chorus/louder its missing the 6-8khz tack sound which is really prominent on most of the tracking that was done.
Funnily enough I can hear a lot more of the players hand movements which is kind of interesting…could prob gain those up because its really authentic and I feel like Im there right in front.
Here is the LALAL version…its not night and day but it def has a different…duller character but compared to the unmix song of SL its def better…what was your workflow, is it actually from SL (or logic pro? as that uses 3rd party as well right?)
As mentioned in other reply, I just want to make music, the LALAL versions seem to work well with non artificial, acoustically tracked sources…
