deMIX Pro is pretty damn good at vocal separations! Check this out

[ar surround]

Not every audio track results in spectacular results. As will all upmix technologies, your results will depend a lot on the source file. ...and some were good and some were not good at all.

Very true. For you all considering this purchase, don't expect miracles. After spitting into multiple vocals, plan to spend significant time cleaning up the stems:

- Clean up of artifacts
- Perhaps taking a few notes from one stem and moving it to the other
- Mixing the vocal stems together into a center image to offset potentially "freaky" sounding separation between the voices.
- Using "Remove Center" or other DAW audio manipulations to get unwanted info out of stems

Be persistent!


The Beatles I Saw Here Standing There with vocals as-found vs. corrected:

Vocals somehow managed to get into the "Other" (guitars) stem. I corrected it by isolating the problem area in the CORRECTED lead vocals stem and doing a "Remove Center" maneuver in Audacity to get rid of it:

 

[JonUrban]

'ar surround' is right. If you want (close) to perfection, in a lot of cases it will take meticulous patching of audio from one channel to another, doing a 'mix paste' or overwrite. If you've ever done that before, you know it can be frustrating and a real pain in the butt.

I only do it if it's a song a really am attached to

 

[J. PUPSTER]

I'm used to working with RX 8 "Music Rebalance" for my stem extractions; so I thought I'd try a little shoot-out between it and the new DeMIX Pro.

here's a segment of an up-mix I'm working on with just Vocal extraction- The top Fronts only from a Penteo 4.1 with the DeMIX Pro on the top screen shot and RX 8 on the bottom. The DeMIX Pro seems smoother and definitely less artifacts between vocal segments. And the second screen shot (Audacity) you can definitely tell the extra garbage artifacts in the Spectrograms.

 

[JonUrban]

Pupster is ALL-IN!! I love it!

 

[barfle]

Maybe this isn't quite the same thing, but I scanned an article from the October 2022 issue of IEEE Spectrum that covers a similar system. Unfortunately, it talks a lot about "AI" but doesn't really say how it works. Anyway, if you are that deep in the weeds, you might like it.

 

[ar surround]

'ar surround' is right. If you want (close) to perfection, in a lot of cases it will take meticulous patching of audio from one channel to another, doing a 'mix paste' or overwrite. If you've ever done that before, you know it can be frustrating and a real pain in the butt.

I only do it if it's a song a really am attached to

Amen to that, Jon. You've got to be attached to a song to work on it. In posts #22 and #23, I stated the problems that I had separating the vocal parts on Twist and Shout. But I went back yesterday and spent quite a number of hours digging deep into the stems and came up with something that shows real separation of the voices between the front speakers.

[DELETED]

It took 8 channels to create these front vocals. I was actually able to separate out the distortion in Lennon's voice, but the result was too "sterile" and just didn't sound like Lennon. So I added the distortion back into the mix. (It looks like Giles was correct to go back and dirty up the sound of The White Album a bit.)

Here's what the 8 vocal tracks look like graphically. Tracks 7 and 8 are a mono combination of Tracks 5 and 6 which are the backing vocals. The purpose of Tracks 7 and 8 is to adjust the spread of the backing vocals. Too much spread can get freaky sounding and also exacerbate artifacts from the cut and paste process.

 

[JonUrban]

Amen to that, Jon. You've got to be attached to a song to work on it. In posts #22 and #23, I stated the problems that I had separating the vocal parts on Twist and Shout. But I went back yesterday and spent quite a number of hours digging deep into the stems and came up with something that shows real separation of the voices between the front speakers. Have a listen to this mp3 sample:

View attachment 85639

It took 8 channels to create these front vocals. I was actually able to separate out the distortion in Lennon's voice, but the result was too "sterile" and just didn't sound like Lennon. So I added the distortion back into the mix. (It looks like Giles was correct to go back and dirty up the sound of The White Album a bit.)

Here's what the 8 vocal tracks look like graphically. Tracks 7 and 8 are a mono combination of Tracks 5 and 6 which are the backing vocals. The purpose of Tracks 7 and 8 is to adjust the spread of the backing vocals. Too much spread can get freaky sounding and also exacerbate artifacts from the cut and paste process.

View attachment 85640

Holy crap, John, you have a lot more patience than I do. But your results are amazing!

 

[ar surround]

I came up with a neat trick tonight for working around a problematic vocal separation. If the track starts with a solo voice and then massed vocals come in later in the song, put about a 10 second segment containing the massed vocals at the beginning of the song. The algorithm will recognize the massed voices early on and have an easier time separating the voices throughout the track.

 

[Charly Mann]

I came up with a neat trick tonight for working around a problematic vocal separation. If the track starts with a solo voice and then massed vocals come in later in the song, put about a 10 second segment containing the massed vocals at the beginning of the song. The algorithm will recognize the massed voices early on and have an easier time separating the voices throughout the track.

That is a great trick. Can you tell me some songs you did that with? Do you have any other software you use for cleaning-up the stems - especially the vocal ones.

 

[ar surround]

That is a great trick. Can you tell me some songs you did that with? Do you have any other software you use for cleaning-up the stems - especially the vocal ones.

I've only done it so far with one song. As noted throughout this thread, I am working on the Please Please Me album. The song I just tried it on is Boys. For whatever reason, the first couple of passes-- once using Local separation and once using Cloud separation-- would not separate Ringo's lead from the backing vocals in several places. I couldn't understand why the algorithm would get confused because there is a major variance between the lead and backing vocals. That's when I tried the above trick.

The only other software that I am using is Audacity. I use that to clean up the stems where appropriate.

 

[JonUrban]

I came up with a neat trick tonight for working around a problematic vocal separation. If the track starts with a solo voice and then massed vocals come in later in the song, put about a 10 second segment containing the massed vocals at the beginning of the song. The algorithm will recognize the massed voices early on and have an easier time separating the voices throughout the track.

What a great idea. Makes sense too. Nice job

 

[Charly Mann]

What a great idea. Makes sense too. Nice job

You are very clever on figuring this out. I use AUDACITY too - but also SOUNDFORGE 16 and Spectralayers 9. Are you mixing PLEASE, PLEASE ME as 5.1 or stereo? If you are doing 5.1 how are you creating the LFE channel for these songs?

 

[ar surround]

I've only done it so far with one song. As noted throughout this thread, I am working on the Please Please Me album. The song I just tried it on is XXXXXXX. For whatever reason, the first couple of passes-- once using Local separation and once using Cloud separation-- would not separate Ringo's lead from the backing vocals in several places. I couldn't understand why the algorithm would get confused because there is a major variance between the lead and backing vocals. That's when I tried the above trick.

The only other software that I am using is Audacity. I use that to clean up the stems where appropriate.

Here is a MP3 snippet of it mixed down from 4.0 to stereo. Nice separation of the vocals. (The backing vocals are slightly emphasized for demonstration purposes)

[DELETED]

 

[JonUrban]

You are very clever on figuring this out. I use AUDACITY too - but also SOUNDFORGE 16 and Spectralayers 9. Are you mixing PLEASE, PLEASE ME as 5.1 or stereo? If you are doing 5.1 how are you creating the LFE channel for these songs?

Charley,

How do you like SF16. I upgraded my version, as I have always done since Version 4.5, and I HATE IT.

I mean, it works fine in every respect, but they removed the ability to select multiple tracks by holding down the CTRL key while you click on tracks. You can select one, but no longer more than one. I HATE THAT as it makes it annoying when you are editing 5.1 or more channel files. I complained and they said they'd fix it, but I just downloaded the latest patch and it still does not work like the older versions.

I am back to using SF15, which tends to crash on me quite a bit

 

[ar surround]

You are very clever on figuring this out. I use AUDACITY too - but also SOUNDFORGE 16 and Spectralayers 9. Are you mixing PLEASE, PLEASE ME as 5.1 or stereo? If you are doing 5.1 how are you creating the LFE channel for these songs?

Please Please Me is being mixed to 5.1. Some songs have a silent Center and LFE while others have one or both of those channels active. It depends on how the initial mix sounds....and if I need the center to fool with the spread of the center vocals or mask artifacts. I really hate the center channel because it is groomed to accommodate my center speaker but may not sound that good on another system. But sometimes it is necessary to lock-in the lead vocal. Regarding handling the bass and LFE, I am only using the left side of the separated stereo stem. I 'double mono' it, and that becomes the centered mono bass for the track. If the base sounds good, I leave the LFE silent. I'll add a little bit to the LFE if I think it may assist with the overall tonal balance.

BTW, I did a separation of the guitars on Ask Me Why. DeMiX Pro separated the electric guitars from the acoustic guitars nicely.

 

[Charly Mann]

Charley,

How do you like SF16. I upgraded my version, as I have always done since Version 4.5, and I HATE IT.

I mean, it works fine in every respect, but they removed the ability to select multiple tracks by holding down the CTRL key while you click on tracks. You can select one, but no longer more than one. I HATE THAT as it makes it annoying when you are editing 5.1 or more channel files. I complained and they said they'd fix it, but I just downloaded the latest patch and it still does not work like the older versions.

I am back to using SF15, which tends to crash on me quite a bit

 

[Charly Mann]

I agree with you totally about SoundForge 16. I actually use version 14 now as my standard because of the problems for me with both 15 and 16. You are really the master of DEMIX PRO 4 editing. I love the samples you are posting. Do you ever share full tracks somewhere else? I think the AUDIOSOURCERE people should pay you to do a tutorial video on YouTube that includes some of your "tricks".

 

[Charly Mann]

Here is a MP3 snippet of it mixed down from 4.0 to stereo. Nice separation of the vocals. (The backing vocals are slightly emphasized for demonstration purposes)

View attachment 85735

This is incredible. Giles Martin and Peter Jackson would be impressed.

 

[ar surround]

Here is and MP3 snippet of a 2.0 mixdown from 5.1 of XXXXXXXXXX. Note the separation of XXXX XXXXX and XXXXXX voices.

[DELETED]

Here is what the completed 5.1 looks like.

 

[ar surround]

Here's a test of separation of acoustic guitars and strings from each other. Both MP3 snippets are extracted from an already extracted "other" stem.

[DELETED]

[DELETED]


If you listen carefully, you will detect minor artifacts of the strings still remaining in the acoustic guitars stem. I suspect that these won't be noticeable when reassembled into a full 5.1 mix.