Topic: AI stems separation: piano vs. strings or orchestra

[hoserama]

I'd be more than willing to do a test run for Logic Pro models if you'd like to compare.

[WiFiJeff]

I'd be more than willing to do a test run for Logic Pro models if you'd like to compare.

pm sent

[phil_er_up]

In SL there is a "Unmix song" and there is one layer for "Sax and Brass" maybe should of run that for you. Put the original file from wifijeff into wavelab with the piano stem and it brought out the piano and he could adjust to his liking.

IMO RX is way behind the rest of the stem software. It does not sound as good as other stem software and has a metallic sound to the stems. The interface is clunky. You have to save each stem separately after it splits the original file into stems. They did not add an automatic stem saving wav files...arg. RX is the slowest processing between UVR and SL. Hopefully they step up their game.

Thanks to hoserama for the info about 24 bit in UVR. Where did you see that option? Maybe I have an older version of UVR.

hoserama I would be interested in the Logic pro models stems if you run them for Wifijeff. Will send you a PM.

[robgronotte]

Nvidia - the AI stemming models use the Nvidia CUDA toolkit. So you'll need a compatible nvidia card to supercharge the processing...makes a huge difference in terms of timing. 2 hour show for the htdemucs_ft model typically took 8-9 hours when I was just running on CPU...takes 20 minutes with my Nvidia card.

16-bit/24-bit - My UVR has a setting to allow for 24 bit.

Logic - Buddy of mine has a beta program where he extracted the Logic Pro models (and a few others) and we've been running it in windows. Still in development but it can be done outside of the mac environment. I've stopped using demucs ever since getting the LogicP models running. I'll make a post on here when it's ready for primetime.

UVR5 Can do 24 bits, but only 44.1kHz, which no one really uses except as 16 bit.

[hoserama]

If I recall correctly, the demucs models were only trained on 16/44.1k. So it can scale up for bit rate but not sample rate, which would carry over for UVR as a front-end for demucs.

Sent the Logic Pro stems over to WiFiJeff, feel free to share!

[hoserama]

Thanks to hoserama for the info about 24 bit in UVR. Where did you see that option? Maybe I have an older version of UVR.

Pulled up my UVR. If you go to Settings > Additional Settings > Wav Type, it'll let you select the various sample rates. I have the most current version of UVR, but it's been included in there for at least a year that I've been using it.

[AbbyTaper]

Thanks to hoserama for the info about 24 bit in UVR. Where did you see that option? Maybe I have an older version of UVR.

Pulled up my UVR. If you go to Settings > Additional Settings > Wav Type, it'll let you select the various sample rates. I have the most current version of UVR, but it's been included in there for at least a year that I've been using it.

I've got the latest version, and I don't see a "Settings" menu

[one8ung]

Thanks to hoserama for the info about 24 bit in UVR. Where did you see that option? Maybe I have an older version of UVR.

Pulled up my UVR. If you go to Settings > Additional Settings > Wav Type, it'll let you select the various sample rates. I have the most current version of UVR, but it's been included in there for at least a year that I've been using it.

I've got the latest version, and I don't see a "Settings" menu

Hit the wrench next to the start processing button.

[AbbyTaper]

Thanks to hoserama for the info about 24 bit in UVR. Where did you see that option? Maybe I have an older version of UVR.

Pulled up my UVR. If you go to Settings > Additional Settings > Wav Type, it'll let you select the various sample rates. I have the most current version of UVR, but it's been included in there for at least a year that I've been using it.

I've got the latest version, and I don't see a "Settings" menu

Hit the wrench next to the start processing button.

Thanks!

[WiFiJeff]

Update on this: the Logic Pro piano stems were so impressive I ordered a Mac Mini.  I underwent a week of torture as UPS, understandably requiring a signature for delivery, kept changing delivery times, arriving after promissed ends of windows, lost the package once or twice, etc.  Set it up and trained myself to do the minimum I needed in order to do stem splitting.

1) on the original file I was dealing with, the piano stem extracted by the M4 chip on the Mac Mini took only 5-10 minutes for a 42 minute piece, and seemed clearer and better even than the non-Apple hack of the stem extractor run.   I also ran the iZotope Azimuth module to slightly center the piano, which seemed to help a bit more tha the EQing I had done on the first stem I was sent.  Not sure this is optimal, but I am still not enough of a master of this software to be able to do next order improvements.

2) the idea of using iZotope's Azimuth module on the piano stem sent me back to a 2005 tape which I asked about on TS several years ago.  I was ideally then placed for the singers, but the piano was way over to the left of the audience and not at all to the front.  Earlier attempts to use delays, level changes, and Azimuth had the drawback of working on the voices as well, the result was a bit more listenable in parts but not natural.  Short passages sounded good but not the long opera.  Running Azimuth on the piano part alone (using the Azimuth "suggest" button to scan the whole file and set parameters), then mixing that back in really fixed up that tape!

3) a recent piano concerto taped from front center also got fixed by separating the piano out, raising the orchestra by 6 dB, and remixing. 

Voices and orchestra will be the next test.  Meanwhile, the newer version, Logic Pro 12, just droppped, I am waiting for  reviews to tell me it is safe to upgrade.  I see no mention of any big improvements to stem separation, that would motivate taking a risk (it's a free upgrade, but I don't wnt to lose any of what I just got!).

[Billy Mumphrey]

You really shouldn't have posted this, I have to get a Mac Mini now haha :/

Let us know what you hear about 12....