will we see SL 13 in a couple of months???
with very good improvements??
argue!
My money is on Wednesday, the 1st of July.
I agree with you. But I’ll say mid-June (second half).
All this stem-separation nonsense will be dropped, because (a) there’s no way to keep up, in terms of quality, with the online separation services and (b) it has been a distraction from the core function of spectral editing, and worked to the detriment of SpectraLayers as a product.
It is being judged largely on it’s ML-based sound-separation capabilities, while its primary functionality is ignored.
But it would be interesting if SpectraLayers 13 includes the new local MVSEP model that can separate audio into 53 (!) different group and instrument stems, if they are allowed to do it under the current license conditions (I think it’s MIT licensed, so it might be possible). I tested the model with ZFTurbo stem separation and it works good. But it’s complicated to set up if you don’t know Python.
I think that SpectraLayers then also would be the first commercial application to include this model.
Of course no one knows better what to do in this case than the developer @Robin_Lobel , but I too hope SL jumps off this wagon and keeps the focus on audio restoration and spectral editing, specially for post-production.
I believe that ML separation capabilities are important to have, but I think the focus should be on SPEECH mainly, followed by further development of the other modes we have already, such as components. This unrelenting race of music stem separation will result in the ultimate demise of SL if it continues the way it is.
Meanwhile very important modules have been dragging behind, like Declick, Dereverb, Unmix Multiple Voices (which for me at least are completely useless, unfortunately).
Spectralayers is by far the most powerful and capable audio restoration suite in the market today (although many people don’t know this yet), and if Robin is smart about this, he can make it the absolute market leader and the standard for post-production.
Personally, I have zero interest in an online stem separation service because that will inevitably be a subscription model. I avoid subscriptions every chance I get. We are approaching asymptotically what is possible with stem separation, so there is no reason to believe SL cannot offer a very competitive unmixing function. I find it very useful as it is.
SL’s stem separation is indeed already useful, particularly in Cubase via ARA, my point though is that it has become an arms race, where every few months a better model is trained, and with an annual release cycle there is no way SL can keep up with that.
Agreed, but for now, it’s free, so why not avail of it, and conversely, why should SL development resources be wasted in trying to keep up with it?
A nice compromise would be if SL allowed download and integration of existing (and rapidly evolving) models along the lines of UVR.
I don’t really agree with that.
Certainly an online delivery model has an advantage at the leading edge. But this is an asymptotic convergence. The marginal gains are very small now. So even if SL is a year behind, that is not a huge problem in most cases. Conversely, the 0.001% improvement I might get by subscribing to the latest online service will probably not be an attractive option for me. Maybe for others.
It would be nice if SL would allow us to import custom models we can download from GitHub or Huggingface. There are new releases and better stem separation models every month. Then we don’t need to wait a year until these models get implemented.
they must include the .wav to midi process in their next version RIPX has it and is more beneficial for orchestrators like me to see the notes in DORICO and or the new score editor in Cubase 15 pro to edit the tracks the notes on the stave line..
Alan Russell
I do not agree at all…many other softwares offer audio to midi. IINM
ie Cubase 15 and SLP should work well together
I have Cubase 15 and SLP, but am busy using a different DAW and haven’t even gotten into the Cubase score editor beyond seeing if I could import my Dorico 6 jobs in to Cubase 15…(which has worked for me)…but I certainly have not tried to convert audio to midi in CuBase 15 nor have tried to work in CuBase 15 score editor as I had done in Dorico.
The more efficiently SLP is optimized for its original use, audio post, the better. Let’s get the basics great first, please, rather than add more bells and whistles.
I’d much rather have hot buttons for my preferred brush settings per tool built into the UI rather than key commands only
This is already possible to some extent from within Cubase by adding the SpectraLayers extension to an audio event, doing the stem separation there, and dragging and dropping individual layers from SL back onto the Cubase project window, and finally converting to MIDI within Cubase using e.g VariAudio, or just dragging and dropping an audio track onto an empty MIDI track.
The issue is the quality of the audio-to-MIDI conversion; RipX is not great at that either. A streamlining of that whole process would be a useful additional feature though, the question is whether the function should be done by Cubase or by SpectraLayers.
there are times when i have had an older pre 2010 score of audio files that I need covert to midi for a client. This is the only time it would come in handy. As for the present sprectrallayers pro is excellent.
on another note:
when steinberg finds the way to run Dorico with all your vst instruments seemlessly, it will sell many more products of Cubase and Dorico I want to hear my original instruments in the score for further editing. This is just me only and my work flow All others have their own work flow which I truly respect since 1968
@Alan-Russell There is a big difference between an audio recording converted to MIDI imported into a score editing software and a score that was manually created to be easily readable. There still might be a lot of manual editing necessary after you import the MIDI.
The best audio-to-MIDI converter I ever tried was a piano converter from Bytedance (link to GitHub). It converts piano recordings about 95% correctly, including sustain pedal, which is a massive improvement over anything else I tried (tools like Melodyne, RipX, NeuralNote, Aurally Sound Prism). Sadly, it’s only optimized for piano.
I’ve been an Arranger since 1968-Thus far, I have mastered my skills and always learn daily.
OK, I’m back working in SLP today…
My desire is for Unmix Noisy Speech to be (much) better at separating inhales/ exhales, breath noises, chuckles, laughter
These components very often get sorted into the noise layer
Yes, users can finish the separation themselves…and cover holes with clone stamp and/ or heal
+1 nosisy speech
Seconded!
About stem separation…
If I have to separate stems—which in my case is very rare or doesn’t demand pristine quality—I wouldn’t use online services due to subscription models or their terms of service. Especially regarding how they can use your files for their own purposes…
@Robin_Lobel
For German voices from time to time I get too much of the sibilants “Zischlaute” like eS, eF, Ts sorted into the noise layer, which makes a speaker sound like they lisp.
I then have to skim through all noise layers and cut the sound back to the speech layer.
How about having a more fine-grained Unmix with a separate voice-sfx layer and also the ability to set a - maybe dynamic - threshold for sibilants?