Mono "Love Me Do"- demixed/remixed-Nov 2 2023

Double-A side release next week, Nov 2, 2023.

Includes the “Now and Then” track that was up against the we-can’t-demix-this-well-enough situation in the 1990s.

For those of us who demix stuff, I suppose “Love Me Do” will be an interesting view into results of Peter Jackson’s demix team and then Giles approach for the first-ever LMD stereo mix

You’d think they use spleeter/demucs like everyone else on the planet. Whatever secret tweaks they possess…I dunno.

At any rate, one could…this week…if interested…do a complete demix of the original mono “Love Me Do” to compare to the next week release.

Doesn’t seem to be overly dense…sidestick/bassdrum/hh/tamb/crash&ride cymbal sporadically/harmonica/John&Paul in various unison-harmony moves/bass guitar…there ARE two acoustic guitars (John&George)…which I’m guessing will remain as a mono clump…but hey, you never know.

Pretty sure the Paul and John vocals were approached by Peter Jackson’s guys to completely demix from each other for placement on separate mono tracks…not to mention the harmonica…at least I’m guessing how I’d approach it for more manageability.

At the original session, John & Paul sang the entire lyric as one mono pass into one mic…including John switching to harmonica during the pass.

I guess one could move deeper as well, separating out the EMI chamber reverb.

At any rate, it’ll be interesting.

If you’re talking about this version or this version then they sound fairly simple to unmix to me and I am confident that I could do unmix the entire audio in less time(less than a week) and I’ve also found a new technique/way of unmixing vocals(soprano, alto, baritone/triplets/quartets) in a more efficient way. Unfortunately, this is extremely time consuming and I have other obligations.

However (who knows) if somebody(like the main developer of spectralayers) is interested in me unmixing this then I wouldnt mind doing it.

1 Like

The point of my post…let’s all see what the PJ/Emi team release next week :slight_smile:

And then, each of us (the 5000-6000 guys on the planet doing demix/remix work as biz or hobby) can discuss if “we” think we could’ve done as well via Spectralayers/Cubendo etc.

Me personally…I’m interested in how the mixed guitars will sound in the new, upcoming PJ/Emi project mix.

I constantly try to get a mental handle on if PJ has access to better tools than I use for demixing my own commercial material :slight_smile:

1 Like

Jackson (I think the main AI demix specialist guy on the Beatles stuff is De La Ray) are supposedly using a US Police demixing solution. A couple of friends deeply in the training side said they think it’s probably just a tweaked dataset and training using the usual approaches of MDX, Demux, GSEP, Ripple, VR etc.
Have to remember they had the 3-Track mastertape to demix the last album so usually only actually 2-stem demixing so possible to get very good results workable into an Atmos or Stereo remastered quality good-enough for release.

A little pre-release thing-


from a review in Variety…"and there are still five early albums due to receive the full deluxe-edition treatment that we saw with [last year’s “Revolver” redux. And beyond that, it seems inevitable that the group’s scream-saturated live recordings and countless hours of video footage will get an AI treatment as well. Beatles revivals will outlive us all…


Believe-it-or-not, I just saw saw the new thumbnail for this on disney+ last night and was tempted to watch but honestly (and maybe it’s just me) I’m not really interested in the beetles. When I saw the thumbnail for this last night and saw that it was a newly added show I actually thought about this thread and said to myself “ohhh that’s what @DosWasBest is talking about”.

I still haven’t watched the first one (the one with “Peter Jackson”) because it’s not something that I’m interested in. In order for me to watch these docu-series I have to really pull myself together to focus in on it (because by nature it’s not something that I’m naturally interested in). I am (however) fascinated with the previews I was able to see so far of them reviving old footage (which would be my sole reason that would drive me to watch this), and (honestly speaking) I’m actually impressed with how “Peter Jackson” was able to not only revive the footage but upscale it and remaster it and bring it up to theatrical digital standard.

Judging by the amount of comments and traction these series has gotten, I will definitely be checking this out, I just have to get in the right frame of mind to focus in on this and give my full attention to this when watching (otherwise I will probably fall asleep watching this).

I found another post (from another forum) relating to The Beatles and A.I. and figured I’d share the link here. Some interesting conversations taking place on that thread/topic

And now, as of today, the red/blue albums are out.

Several demixed/remixed tracks among the collection.

Until today, I didn’t really pick up on the “Mal” term (nods to Hal in 2001 Space Odyssey and Mal Evans).

So, apparently Jackson has more…or better…or arguably different training models than everyone else on the planet. Which I don’t necessarily believe (interesting what one person mentioned about police models or whatever).

I read Robin saying somewhere that there may be a limit to the number of models that can be developed …by mankind…
due to infinite combinations of stuff in mankind’s total recorded output.

I dunno. I don’t see how there can be an outer limit given enough time.

Plus…and I’m not a programmer…if spleeter, demucs etc exist now…and did not exist in 1975…seems to me that something will eventually come along as core code that will dwarf what current code can do here at the edge of 2024.

Which of course…will have the planet starting all over on remix projects :slight_smile:

I don’t remember saying something about an upper limit.
It’s unlikely Jackson has better models that the state of the art AI these days. However he has access to the original Beatles stems, which makes him able to train AI models specifically for that use case (and it’s likely those Beatles models would perform poor on any modern recording for instance - it’s really a matter of focusing the training for a specific use case).

About the topic of “limitations”:

I believe that is in-regards to ethics and pending regulations within the company’s producing these A.I. models. Before all these A.I. art generators started popping up, there were already internal talks about banning celebrities or people being used within their input inquiries (out of fear of lawsuits and being sued).