New Stem Separation Software & 2 Beatles Stereo conversions using it!

[Mixerrog]

All this AI stem separation software actually started with the first version of Open-Unmix that I started using about 2 years ago. Spleeter & several others are based on this type of AI teaching for a training table & finally Open-Upmix issued a new UMX version that has a added to the training table that now actually includes a pre-trained speech enhancement model umxse provided by Sony even!

It took me a few hours to get the dos software going but I finally did + with a batch file now, it is very easy to run! You can read about it at https://github.com/sigsep/open-unmix-pytorch The stem separations are 32 bit best quality so far I have heard from any AI software so it will make for some great new mixes IMO.

I am finding the vocal & high frequencies stem separations are much better than any previous software. There is just over 15 khz cutoff on these stems as the new training files had the 15 khz cutoff but this is still better than Spleeter & most others that actually use a 11 khz training table & then digitally interpolate to 16 khz! This is why the higher frequencies sound much better than others IMO.

I used this new program with a little spectral cleanup to produce new stereo conversions of I'll Get You & P.S. I Love You as examples of how well the program works the you can get for 1 week with password of 092849 at hxxps://we.tl/t-6Ewy0fJ4YE

Enjoy, Rog

[miggynav]

Thanks! really great separation and mixes.

[MP3J]

i'd love to use something like this, but i'm just not happy without a simple GUI.
I'll wait until somebody clever developes one .

[Cranberry Sauce]

Both of these sound great. Thank you.

[ToddS]

Very Good . Thanks for always sharing your audio with us here and at BSN.

[Tex]

I am finding the vocal & high frequencies stem separations are much better than any previous software.

I'm not an expert but it seems to me it's the high frequencies (and high frequency reverb) that really gives the perception of genuine separation.

I mean almost all of these spectral mixes *sound* like stereo but they don't *feel* like stereo if that makes sense.

Hopefully this new software is a big improvement and can be made usable to a non-technical person.

[Mixerrog]

Engonoceras,

The training table for this new UMX software is 15.5 khz so cutoff is there but Spleeter actually uses a 11 khz cutoff table with digital interpolation above that to 16 khz so the high frequency separation & sound quality is definitely superior to Spleeter & most other A1 software as it also has a 400 mb larger training table which may not sound like much but that is double what Spleeter & most others use including iZotope RX8 which means better separations also.

Rog

[Sologui]

Thank you for this info, AI technology is progressing, more and more people are interested in it ... One day we will be able to create new Fab Four songs, just by AI intelligence (There are already working groups under Python working on the subject)

I tried, in the past, to install Open-Unmix, I did not succeed, but I did not insist either. I will try again

Thanks for sharing these two stereo tracks, excellent!

So that we really realize the effectiveness of the Open-Unmix program, could you share with us the isolated stems of a song?
Thanks in advance

[Mixerrog]

Sologui ,


You can grab the 32 bit stems of P.S. I Love You using the new Open-Unmix UMX at hxxps://we.tl/t-iHMP5zF0Z3 usinf password of 092849 for 1 week.

Rog

[Mixerrog]

I got the 24 bit flac output version going & it is at least twice as fast so I am converting to it.

You need to install "stempeg" first by running pip install "stempeg" from your first command prompt.

Then run --

umx --ext .flac --audio-backend stempeg "input song" --outdir "directory location for flac output stems"

Rog