for music separation, generation and remixing (oh, and automatic text

https://www.musicradar.com/news/audacity-free-ai-plugins

This seems like a no brainer to download and install.

https://github.com/intel/openvino-plugins-ai-audacity/releases

And it was. Took about 3 minutes to first view download and install video, then download, extract, copy and paste. And all was done.

Thanks for the tip pghboemike.

This is great news. I already use Audacity in my workflow and it will be awesome to use their AI tools. I sincerely hope Audacity can pave the way for using all sorts of AI in music production.

pghboemike, thanks for bringing this to our attention.

I agree with Joanne, it would be great if Audacity could pave the way with AI-driven editing tools that really work in a solid way. Audacity was my first audio editor and recording application. I'm skeptical though, especially with stem separation in music with synths and lots of reverb. But, someone will eventually get there, we're still in the early days of this technology.

I just a made a quick test. Easy to install but took very long time to run on my PC to extract the stems for one file that I tested.

I still think the quality of the extracted stems varies between the various software products, I just made a quick test between Audacity, Steinberg Spectralayer 10 pro and SongMaster pro and Moises ( Moises m4a conv to wav in SO). I extracted the stems from a common country song in mp3 and extracted it as wav. Had a look at and compared the spectrums with Audacity spectrum analyzer for the four different extracted Bass Stems. They had significant different looks, see attached jpgs below.

I wonder if anyone else have any experience regarding the quality of the extracted stems from the various tools available.

Your thinking on this is absolutely spot on. There are so many tools out there that claim to do stem extraction (and they probably all do to some extent) but I've read reviews on such tools and I can't find a review that actually posts audio results.

Questions I have for a given tool are:
What is the resulting sonic quality of the extracted stems? How much background noise and how many notes or chords were missed in the extraction?
What were the specific stems extracted?
In what genres does the tool fall down on?
In what genres does the tool really shine?

I wonder if there are any benchmark songs that have been established across genres such as progressive rock, country, smooth jazz, bluegrass, etc. to help evaluate the current tools in the marketplace. I'm thinking that only a handful of carefully chosen songs would be needed to determine if tool xyz would be a useful fit for a given muscian.

I tried Spleeter, Moises and Demucs some time ago and made a quick comparison:

Wow!
As this thread develops it's getting better and better.
Great job Cerio!!

Do you have the time to do the same for A Cat On The Chimney but extract the piano and maybe the bass?

Cerio,
Thanks for informing me about Spleeter.
Here is the piano stem from that program. This is what I suspected, the quality is just not there. To be sure, this is a tough technical nut to crack, and they have made great strides but (at least for Spleeter) they need to do more development work for piano stems to be useful to me. The bass stem is better, and could be useful. Bass is easier because of where it sits in the spectrum.

I did the same with Song Master Pro and got essentially the same results.
I wonder how Audacity or other programs do on this song.

Spleeter Piano Stem

Most of the existing stem separation algorithms (many of which use actually spleeter, with custom training models) do already a good job separating bass, vocals, drums, and the rest from a mix. I guess it will be a while before these algorithms will be able to process guitar and piano tracks with acceptable quality.

Stem separation is a new technology, but it's advancing really quickly.

Totally agree.
I learned a few things here, thanks.

Has anybody got the music generation thing to work in Audacity? Trying to find out what it does but keep getting errors.

I bought Band in a BOX ages ago for my Atari Computer. I wanted to have backing tracks to sing along.
For the same reason, I now try to get karaoke style backing tracks from these new AI tools for stem separation.
And I tried a lot, even the iZOTOPE product did not produce useful results (some say that these are based on spleeter but have been enhanced with better training).
It does not life up to my expecations. Even if I just want to have the voice removed, the resulting stems lack some of the Instrument parts, but some artifacts from the voice are still present. I could get rid of some artifacts using melodyne to clean things, but that is to much work for to poor a result.
And the stem separation for Audacity is no exeption, I just tried this and it is nor better than worse as all the others.

Maybe AI for music is getting smarter over time.

I think SongMaster Pro works fine for producing backing tracks to sing along to (at least it works fine for me).
Just drag in the song and split up the stems in the mixer and just mute the vocal.

Just tested the Music Separation in Audacity (vocal, and the rest) it seems to work fine as well.
But SongMaster Pro is so much more and makes singing to the extracted mixed backing track so much easier with the options to add in the lyrics at the exact timing locations. As well as showing the lyrics in the time line, could also as well be shown in a lyrics window or karaoke window (trial version is available).

I have worked with a couple of these stem removing tools now and find then fantastic (and generally equivalent for what I need). I compare them to what we could do a couple decades ago and consider them a modern marvel. Game changing!

I am just finishing up a project where I do a cover of Goodbye My Lover by James Blunt. I used only the piano from a live performance and all the rest I added. Could not be happier.

It worked as claimed. I selected GPU option which was default. It appeared to run in real time since it took as long to process as the song was. It does the stardard 4 tracks: Drums, Bass, Vocal, other instruments. What more can you ask given it is free!

Hello Joanne,
I have got music generation running.
I had errors first, but when I put all load on the CPU (some tasks are set up for the GPU by default), it did work.

I guess that is because I have turned off my Intel GPU (that is part of my Intel processor) in BIOS. I use a separate NVIDIA GPU.
By the way, Spleeter makes use of this NVIDIA GPU.

I can imagine that, openVINEO being Intel stuff, it would only work with Intel GPUs.
And even if it does work on Intel GPUs, that may not apply to the GPUs within the processor chips but only with their own Grafik Cards.

Anyhow, I could run it and it did something. I did not give it much input in the promt input window. I guess the more specific you are the more control you have regarding the output. Anyhow - it did work for a while and then I had a wave file that was sitting in an Audacity track that I could play and manipulate with the usual tools this DAW provides.

How to properly prompt this tool I do not know. A manual was nowhere to be found. There are such music generating applications running in browser windows. These work with prompts like : a ballad in the key of C with piano and in the style of ...

Maybe this VINEO stuff accepts the same?

But now that I have seen what it is and does, I see no benefit it has for me. I do have a music generating software that is giving me much more control - Band in a BOX.

Thanks so much Uncle Paul. I will give that a try. I made contact with the product manager for Audacity and am hoping for a meeting with him to discuss a possible collaboration with LyricLab. I am battling to envision how LyricLab can work with Audacity.

I tried AI with Audacity as well, lots of errors, had to give up on it. Thought it might have something to do with my aging computer, but seeing the above seems not the case.

Song master pro works great here for extracting the Stems.

Thanks Uncle Paul. I got it to generate using your advice and switching to advanced. Does anybody have any suggestions on how ai lyric generation could be integrated with Audacity?

Just to be clear, the Spleeter code will not pull the paino out clean. This code is optimized for Bass, Drums, vocals. The piano will fall into the "Others" which is the fourth stem in all these spleeter applications. Use the app on Bass, Drums and Vocals to see how well it performs.

I just did a workaround to get piano. I chose a version of the cover which was Live, so vocal and piano only. So the "other" track was very clean piano.

Two days ago...

v3.5.0-R2 Latest
🎉 We are thrilled to share this new release of OpenVINO™ AI Plugins for Audacity*! It includes a bunch of improvements over the initial release, and is compatible with Audacity 3.5.0! 🤘

💾 Download the installer here

https://github.com/intel/openvino-plugins-ai-audacity/releases

dont wanna start an argy bargy...i dloaded audacity and the open vino installer a few days ago...turns out then i gotta dload models on line ? i assumed such would be included with vino installer.
my problem like lots of studios is pc is not attached to the net.
i use a seperate pc to dload..
you guys must be on line ??

the whole process i found not trivial.

i also wanted to compare audacity as a daw to rb.
Yes there are lots of nice enhancements but i would lose more than a few rb features i like...also...to get some audacity plugins that interested me THEN i gotta download something else...sigh.

guys...in case your wondering i DO try to to be open and unbiased..but i found the whole multiple installer approach to get what i wanted ie the special plugins and stem seperation features...not trivial....and time consuming....
LET ME REITERATE...if i found something better than rb i would switch but so far i havent. also the rb generation and midi features are so extensive and keep me as a user of rb.


i think there should be ONE audacity installer and the whole process simplified for studios not attached to the net....also one download screen where the user
selects optionally the vino and special plugins to include with the audacity installer.
im thinking of new users to music production who would find the whole process complex.

happiness.

om 🇨🇦 🇬🇧

I just double clicked on the downloaded installer file and it installed ok, it was several steps but all automatic, and I was on-line.
But Audacity needs to be 3.5.0 (I had 3.5.1 and had to find the older version). But now it works.