SL12 Pro for Splitting Lead/Rhythm Guitar / Advice and Sourcing Video Tutorials

SpectraLayers
21

I’m using MVSEP on my latest project which is a one hour live project.

First a technical introduction for those who bother to read…

This project was a stereo live recording from 1987. It’s not a good recording by any stretch. It was recorded with two line level microphones, Milab LC-25LL, into a Milab 48volt phantom power, and then into a consumer converter, Sansui PC-X1 and stored on a portable Sony Beta Video Recorder.
AD specs: 44.056kHz, 14 bit; which is actually between 11 to 13 bits on a good day. It was a very poor sound quality – but, that’s what I had…

In the summer of 1998 I copied all my recordings analog to my (then) new Ensoniq Paris before exporting to 44.1kHz, 16bit audio CD for storage

(– and wishing for a technological development, as we have now).

After importing the AIFF files I converted everything to 44.1kHz, 32-bit float, wav files in Cubase. This is also my working format for the entire project, which will later be edited and converted to 48kHz, 24bit master…

– Why 44.1kHz? MVSEP does not deliver higher than this sampling frequency. As far as I understand, it has become a kind of defacto standard on most/all separation models… I first tried to convert to 48kHz along the way, but eventually thought that there was no point - started everything over again in 44.1kHz…

This production was initiated by a local small town choir with a hired baritone vocal soloist and musicians; grand piano, bouzouki/guitar, flute, violin, cello, double bass and tambourine and dancers.
The show was 18 songs from Theodorakis in a dramatized story with ballet dancers.
The set design was; musicians in a small space on the far left of the stage - the choir, the soloist and the dancers moved around as the show required…

This time I have separated all the instruments and vocals with MVSEP, except for the bouzouki and guitar in Logic Pro, and I think everything turned out magically well (under the circumstances).

– Today I discovered that MVSEP had a model that separated vocal solos from choirs. I had no illusions that it would work on such rotten sound (or if it could work at all).
But – it turned out an incredibly good result. I have only separated one song with this model so far. It was extremely impressive.

I do all the editing in SpectraLayers, where I put all the tracks in their own layers in SLP and finish them before exporting them to a transit folder – from where I import into Cubase 14.
I really love editing in SpectraLayers :slight_smile:

The purpose of all this: To be able to offer this show to be heard in a restored and revived experience within a concert staging.
I have received great help from ChatGPT with the scenography; instruments, soloist and choir placement in both width and depth.

– What a lovely hobby… :slight_smile:

I hope this can be of interest to like-minded SpectraLayers users…

2 Likes
22

I also work in 48kHz but honestly given the imperfections inherent in all of these stem separation techniques—whether manual or automagical—the difference between 44.1kHz and 48kHz is probably the last thing you’ll notice when you’re done. Nonetheless seems most of us use at least 48kHz so would be nice if the tools matched up… Steinberg if you’re listening. :slightly_smiling_face:

1 Like
23

Most or all models are trained and tuned at 44.1kHz.
I don’t quite understand what Steinberg can do with this…

1 Like
24

the SL developer sources free stem separation models from the same sources used by MVSEP and UVR

1 Like
25

I mean for training “overall AI”…as a sample of a human person’s music writing to then feed the “AI machine” that is regurgitated the AI generation (generation as in “creation”) of music as induced by prompters using text.

26

awesome!! I shan’t quote all your text which I certainly enjoyed reading. Sounds like a great project.

I can fully relate to this as I have similar, stereo live gig improv recordings to DAT…about 11 hours of it. DAT tapes were expensive in the early ‘90s and so we often recorded at long play which is 32KHz and probably 11bits as well. I think 44.1KHz was only available on pro DAT machines. My Sony DTC-690 was 48/16 at standard…our drummer also had a DTC-690. We recorded plenty of performances at 4816.

Our instrumentation was elect guitar/ Roland Octapad (so synth drums), cello (sometimes played in unorthodox ways, for example dragging a rubber ball on the cello body) or salad bowl with water and contact mic, or a tape head violin a la Laurie Anderson, where the guy playing had made little bows of audio tape connected to cut off coat hangers (audio tape taped to the plastic coat hangers with scotch tape) a tape head was mounted on a plastic toy guitar and amplified.

These atypical sounds just don’t get noticed by unmix models looking for “traditional” instruments in SL, so I just don’t try. Even my various original music projects do not get unmixed very well in SLP. The NR possibilities of SLP, on the other hand, are simply fantastic.

I did most of my transfers from DAT in 1999, so similar to your project timing.

1 Like
27

I work in whatever sample rate the task calls for. Something I noticed with running NR VSTs (Acon Restoration Suite) was the bit depth clearly made a difference when processing the NR. Clear, to me, anyway…maybe I was wrong about that, I don’t have any scientific proof of my claim. It just seemed like some unwanteds would slip through at 16 bit whereas 24 bit seemed to produce superior results. Maybe hat was how my DAW worked? I’m not sure.