Can you isolate guitar, vocal tracks from 1-track recording?

LiveMusic · #1 03-14-2024, 09:02 AM

Say you have a 1-track recording, guitar/vocal. I know there is software that even general public users can separate the vocal from a produced song. So, is it possible to separate a guitar/vocal work tape into two tracks and then you can tweak the levels and e.q. separately to make a better final product? If so, does it work decently? I know, I know, record on two tracks, lol.

Doug Young · #2 03-14-2024, 11:55 AM

There are services and programs that can do that, within limits. I've use https://www.lalal.ai (website) pretty successfully with some tracks. It seems to depend a lot on how clean the recording is.You can try it free to see if it works, then pay if you want to download the full tracks.

FrankHudson · #3 03-14-2024, 02:12 PM

Acon has a software product "Digital Remix" that is under $50.

I've used it some. Results vary, what I've seen is that it will separate pretty well,* but there are some artifacts, little bits of the unwanted signal that will stay in and/or some impact on the wanted signal. Sometimes that is mimimal, and particularly in a mix of several instruments, less noticeable than the leakage signal being left in.

Easy to use. The interface shows vocal, guitar, bass, drums, and other and there are presets to remove, or leave onl, each of those.

I think they have a free trial period, worth checking out for your purposes.

*it's generally very good in separating bass and drums from each other. Often good in separating out slight guitar leakage from a vocal track. Can sometimes work well enough in removing vocal leakage from an acoustic guitar track. Often not that great with misc. instruments, like keys, strings, wind instruments etc.

LiveMusic · #4 03-14-2024, 04:00 PM

Thanks for the posts. I tried the LALAL.AI and it seemed to work perfectly. I paid for a package and then got going. It did its thing for a guitar/vocal and I got the files in a zip file. Unzipped. Imported into Cakewalk Bandlab as tracks 1 and 2. Well, heck, even though if I listen to each track separately, it seems all is well. All I can hear on the guitar track is guitar and all I can hear on the vocal track is vocal. But when I have both tracks loaded and try to pull the guitar down, it pulls down both the guitar and the vocal. Same if I apply the same to the other track. It's like the vocal and guitar have NOT been separated; it's like these two tracks are identical, even though what I hear with my own ears is vocal-only is on the vocal track and guitar-only is on the guitar track.

Only anomaly I am aware of is the Bandlab kept asking me to active and I've done that way back. I even did it again. I got no warnings other than that.

Mind you, I do not know much about recording but that seems a pretty basic move, lol. If there is anything I could be doing wrong, advise, please. I was pretty fired up about the LALA.AI!

Chipotle · #5 03-14-2024, 04:05 PM

Just make sure you don't have something like both tracks selected, so you actually *are* pulling down both tracks at once! I can't think of anything would make tracks that are actually different sound "linked" like that.

dnf777 · #6 03-14-2024, 04:06 PM

Hold on to your hat. When the recording industry adopts AI, things will be possible that we havent imagined yet!

LiveMusic · #7 03-14-2024, 06:41 PM

Quote:

Originally Posted by Chipotle

Just make sure you don't have something like both tracks selected, so you actually *are* pulling down both tracks at once! I can't think of anything would make tracks that are actually different sound "linked" like that.

THANK YOU, I got it fixed. Not sure why it did what it did, lol. If either the "1" or "2" track numbers is selected on the tracks, it causes this problem. Otherwise, deselecting both, it behaves as it should. So, now I can adjust volume and e.q. for the vocal and the guitar separately. This LALAL.AI looks like it works pretty darn well to me (actually REALLY well) but I have limited experience at recording other than recording countless one-track recordings over the years. A few multi-track, lol.

It will be interesting to finish this and see how improved the final is after I used this magic versus the original single track recording. (Problem being, mostly, that the acoustic guitar is too loud.)

LiveMusic · #8 03-16-2024, 09:04 AM

Quote:

Originally Posted by Doug Young

There are services and programs that can do that, within limits. I've use https://www.lalal.ai (website) pretty successfully with some tracks. It seems to depend a lot on how clean the recording is.You can try it free to see if it works, then pay if you want to download the full tracks.

Thank you, Doug, I did use it and I thought it worked great. Except... on one song, it cut off the last two words of the lyric! There was 'space' there to the song's end and no sound on the vocal. I did it over and it did the same thing. Odd! I did a few others and it didn't do that.

A fellow told me that yes, this technology works due to frequency elimination or reduction but when it does that, it also removes some artifacts or other 'stuff' and that can be bad. Sounded perfect to me, lol, but I'm no expert and he is. Although, I do not know if this new crop of "AI" software is better than karaoke vocal remover software of recent years, which seems to be what he was talking about. I don't know if this is better or if he is aware of it. It sure worked for my task.

Doug Young · #9 03-16-2024, 02:41 PM

I believe the old karaoke vocal remover was just a simple trick - since in most mixes, vocals are in the middle, you can play games with nulling with phase to get rid of anything in the middle. Didn't usually work all that well.

I'm not sure what these AI tools are really doing. The software Peter Jackson used on the Beatles was apparently really great, but they had lots of raw data from Beatles recordings to train it with. These generic tools aren't trained on you or your voice, so it's going to be a bit more hit or miss. There's usually a small amount of bleed left, and it can miss some things. I tried it on some really old recordings that were pretty low quality, and it sort of worked, but not well enough to use. New, clean recordings worked great for me.

I'd guess it just hit a spot for you where it got confused. I wonder if you gave them a short snippet, just that section that it choked on, how it would do. Might be possible to fool it somehow.