Hello
I am using the trial version of SL pro 11, for the first time.
My use case is the following.
Band played at a big music festival and the archival video looks great, so I want to post on socials, YT etc,
Board mix was fed to a camera way too hot and is clipped and is beyond repair.
Other cameras have almost useable audio. In one song there is a sax solo that is too low in the mix, I was hoping to use SL to separate the sax solo to a separate track that I can then mix in as needed.
I used the unmix module and selected only “sax and brass”
The resulting tracking is mostly silent during the song - it is not picking up any of the sax.
During the sax solo it separated about 1/2 of the phases from the sax solo, but left the other half of the sax track silent.
What can I do to improve the performance and accuracy of this feature?
I am really hoping I can get this to work as I have a large archive of very well shot footage of the band where the audio was recorded sub optimally with out access to multitrack.
Check all stems. It’s likely the sax part went elsewhere (due to bad recognition, that can happens depending on the sax timbre and recording conditions). I would bet it went to the Vocals part. So unmix with everything checked, and then - if there’s no vocals in your song then you would just have to merge the Vocals and Sax layers, and if there are vocals just use the transfer tool to carefully transfer the sax part (in the vocals layer) back to the sax layer.
yep, gotta roll up the sleeves and get to learning to read the spectrograph and SL selection tools and UI to re-balance what you after
as Robin says, the sax could be spread out to other layers
IME often the “Other” has all kinds sounds that SL isn’t able to identify…I find a lot of fx like echo and reverb finds its way to Other layer as well
Before investing too much time in this ( I am a busy with multiple projects)
I should ask if what I am trying to do is possible.
I assume that the primary use case that the software is designed for is unmixing audio that was properly recorded and mixed ( in studio environments) - where as I am trying to alter live 6 piece band that was sub optimally recorded to a mic built into a professional video camera ( or in some cases an iPhone) . (The second example in the video is rather well recorded)
Suppose I invest a bit of time into learning the software better, how good can the results get with examples like what I am dealing with?
And one more question
The Live track in question is 8 min long and the unmix progress bar is now at 90% - this is after about an hour (at least) on an M1 Studio with 32 GB RAM, I know this is an insane amount of processing going on but is that normal for it to take so long on such a powerful computer?
have a look at @Unmixing video on this forum from about a month ago- (be careful with that vid, there is a pretty serious speaker breaker in the video)
Essentially he is using the Magic Wand tool alongside a few other tricks to work in different channels
Again, I haven’t been unmixing music- I’ve been working on human voice…more than one speaker into the same mics…it generally takes me about 2 weeks to just attempt to get some separation between the human speakers and rebalance levels from maybe a 5 minute excerpt…those separations don’t sound too great solo’d, but when all mixed back together, some impressive results can be achieved
Sax might be a bit easier…if there is only one sax…yet I don’t really know
and realize, it takes plenty of time to get to grips with reading the spectrograph
I’m using SL11 3-7 hrs daily for about four months now…but I only started with it in June 2024…so I’m no expert
I do have some old songs that are live performance with Sax (one sax player)…and the Sax has ended up all over the place in various layers after running Unmix Song…that was with SL10, tho
I hate to point anyone away from SL, yet there are many UVR users who prefer UVR for music instrument separation…nevertheless, once you run UVR unmix algos…you need to reassemble those split tracks somehow and I’d think SL would be the best software to do that…I’m just saying UVR might be worth a shot as it is more running of “modules” than manual unmixing
Download UVR5, then download the VR 17_HP-Wind_Inst-UVR model, and give it a go. For me this has worked substantially better than the SL Sax and Brass separation.
Cheers Al.
Thanks AL808 - never heard of UVR5 but downloading it now, thank you so much for this tip.
ctreitzell thanks for the link to the video , but there are many videos on that page link, please let me know specifically which one you are pointing me to. I am also really curious to know more about your use case with SL11- What are you doing with it that keeps you busy 3-7 hours a day?
Thanks al808 - yes the results on sax are much better but lots of work to line it up with the original track because the UV% solo sax file is a different length than the source material. - What am I missing?
Ok so I tried UVR5 and it was better than SL11 for this job but still has the problem of unequal file length and the beginning and end does not line up with the beginning or end of the source file.
So then I tried RipX and for this task it is the best by far.
I am getting close to a usable mix.
As far as I know Ripx does not have the ability to separate different voices in a vocal track and I now have a 4 bar section of a vocal track that has a male and female singer and the female is too loud, so I brought this (only 4 bar section ) into SL 11 to separate, as I write this it has been going for about 40 min and is at %48 with a spinning beach ball. I must be doing something wrong that SL is so much slower than competitors tools? Any suggestions or settings to tweak or pay attention too?
Hi funkstone,
You’re right about the different lengths.
I did a test with a couple of tracks:
One was 00:08:54.617 and the UVR5 tracks are 00:08:54.607, so there’s a discrepancy of -10 ms if understand correctly. However, I aligned them and the waveforms appear to sample match for the whole file, in other words it seems the process subtracted the 10ms at the end of the file. I didn’t have the line up problem you encountered, a bit of a mystery…
I tested a 00:05:05.482 file, UVR5 gave me 00:05:05.480 , so this time -2 ms, and aligned ok as well.
RipX DAW gave me a 00:05:06.875 file, so it added a generous 1.395 seconds, and aligned ok.
To my ears UVR5 did a cleaner/better job than RipX.
As far as the performance issue with your M1 Studio, I’ve got the basic M1 mini, and the performance is pretty dismal for un-mixing with SL (the gpu does the heavy lifting for stem separation I believe, and Logic’s basic un-mixing is very fast with the M1) . I ended up getting a windows machine with an Nvidia gpu, as most ai-un mixing tools seem to need one.
Also SL is the only one that nulls with the original file.
Cheers.
I have stopped unmixing music for now…I just don’t have time for that at the moment…I, personally, believe the tech will catch up in the next year or two and the time spent wriggling through the rabbit hole of manual unmixing in SL will appear like a waste of time when I look back
for my i7 8700K, 64GB RAM and GeForce 1050Ti it takes 20+ minutes to unmix a 3-4 min song of mainly RnR instrumentation using SL
I’m far too busy with prepping for winter at the ranch and applying SL magic to my documentary
What SL11 excels at is what I’m using it for at the moment: separating noise from speech and then separating unwanted noise from the wanted noise…I know of no better tool for that than SL…and as soon as this documentary is done…another will find its way to my workstation
The thing with unmixing music is that it isn´t perfect and still needs a lot of manual work, if you want top notch results… Woodwind and some brass (and guitar harmonics also) with vocal can be really hard for AI to separate. Also the quality of source matters a lot). There isn´t simple click and done results, and probably won´t be in few years…
