in depth guide to getting cleanest vocals?

[silent tundra]

anywhere where i can find this? techniques for certian situations but usually for rapping / singing models. would apprecaite it.

oh yeah here my work flow on how i be getting vocals and shit :

• invert polarity on audacity
• bs reformer on mvsep.com
• de reverb
• de echo
• de noise
• remove any background adlibs using bve v2

Audacity, noise gate, trunacate silences, then normalize .

[lean bough]

how i clean the fuck of my datasets

• bs roformer/bandit plus isolating
• uvr de-echo-normal/de-echo de-reverb (in some cases)
• rx de-ess, mouth de-click, de-crackle (in some cases), de-plosive
• noise gate, truncate silence and normalize in audacity
• export in flac format (32k, 40k or 48k)
• (in some cases) uvr denoise if audacity and rx aint enough

there you go, my data cleaning lol

[dusky trail]

Just to point out, sounding clean and sounding crispy/detailed are two different things. Your workflow is already okay with it, but if you want crispy and detailed model here's a general workflow

• phase cancel them in premiere pro (advantage is you can adjust the volume automations)
• use either bs roformer or viperx
• use uvr deecho dereverb model
• sum stereo to mono and do spectral editing there
  (lesser workflow, less destructive processing)

tip: never ever use vocals that have backing vocals, forcefully using bve or karaoke models won't work and it what destroys people's models

[silent tundra]

with bve it does remove most of it tho, i have an example where its actually cleaned out double layers of vocals. so there is somewhat a benefit to using it, but i got what you mean

like adlibs too, and usually it doesnt affect quality.

but like said it depends on what it is

[silent tundra]

is premiere pro free btw?

and i thought inversiing or inverting polarity help getting tha process cleaner, since it flips like the stuff

[dusky trail]

not free tho

[dusky trail]

inverting the polarity won't do anything to the quality, it's just the waveform are flipped. But aligning them with full mix and instrumental track and iverting one of the tracks, it cancels the instrumental mix without destroying the vocals

[silent tundra]

so should i keep this in or not do this?

i do it before applying any ai isolation on it

[dusky trail]

if you don't have any instrumental track to phase cancel them, then inverting the polarity of your full mix track is useless

you can do it or not, but the quality is exactly the same

[silent tundra]

ohhh alr

makes sense

[silent tundra]

well thing is i notice a difference doing that

the whole inverting thing

like when i use bs reformer it somewhat takes out adlibs and shit in with the instrumental

lmao

[halcyon dagger]

This is how i clean my e-datasets:

• rx spectral de-noise x2
• voice de-noise (sometimes)
• normalize
• de-ess (only the actually sibilances not the whole dataset)
• de-crackle (sometimes)
• noise gate
• de-click/de-mouth click (on the actually click not the whole dataset)
• resample to 32k
• then a ton of manual

[teal grove]

Just adding onto here; I was told to use EQ if there's dc offset present in the waveform statistics. Helps to remove those low noises on the bottom

[lean bough]

This is how i clean my goofy ahh datasets (updated tho):

• bs roformer/bandit plus, deecho and dereverb in mvsep
• rx spectral de-noise x2
• voice de-noise (sometimes)
• normalize
• de-ess
• de-crackle (sometimes)
• de-click/de-mouth click
• dialogue isolate (sometimes) + de-plosive
• noise gate and truncate silence in audacity
• export flac audio to 32k/40k
  and there u go

[halcyon dagger]

Dialogue isolate can actually outperform bandit plus

[lean bough]

Idk honestly, I use RX 10

[halcyon dagger]

Oh then dialogue isolate is not better