VST Midi humanizer plugin

[adamj]

If you are using Ableton Live with Max for Live, my MIDI Sculptor device may be of interest:

https://github.com/adamjmurray/ableton- ... /README.md

[rbn777]

...

[Kalamata Kid]

If you are using Ableton Live with Max for Live, my MIDI Sculptor device may be of interest:

https://github.com/adamjmurray/ableton- ... /README.md

It's unfair!
M4L gets the best toys.

[rbn777]

...

[adamj]

Any plans for updating this for Ableton 11? Doesn't work anymore, but I really miss this one... Really on point and simple to use
 

@rbn777 / everyone: Sorry, I feel bad. I had Live 11 support well in advance of the Live 11 launch (I was a beta tester) and I never released it. There were a couple bugs holding me back, and then life intruded and I have been on a music-making hiatus for months. Nothing's wrong, I've just had other priorities.

Anyway, thanks for the kind words and the ping. I should have released it because it works fine for the most part. I updated the Github page with a download link for a Live 11 beta version: https://github.com/adamjmurray/ableton- ... /README.md

Or download it directly: https://github.com/adamjmurray/ableton- ... d?raw=true

New features include support for

• Velocity range
• Note probability
• Release velocity

I'd appreciate anyone who tries it and let's me know if you encounter bugs or if it works well.


If anyone's curious, the new MPE support in Live 11 was really problematic because they completely changed the way things work with MIDI and Max for Live. That's why the current version doesn't work at all in Live 11. I had to step back, experiment a lot, and redo the foundation of this device. For example, when you randomize note positions, if they overlap temporarily while you drag the mouse, Live will automatically delete the note. But this is a huge problem because if you keep dragging, the note might not need to be deleted anymore and should re-appear. In Live 11, all the MPE data is lost if the note is deleted and there is no way to see it or restore it from Max for Live. So I had to detect when a note is about to be deleted, and I temporarily shove it into the negative time for the clip at pitch 0. And then if the note gets "undeleted" I move it back. It's a whole layer of additional complexity I had to add to be compatible with MPE, and it kind of feels like a hack job, but I don't see a better way until Max for Live gets full access to MPE data in a MIDI clip.

Anyway... one of the bugs is if you have notes in negative clip time at pitch 0, my device may delete them. I feel strongly my device should never "accidentally" delete a note. But is this really a problem in practice? Probably not. Why would anyone put notes there? That's why I should have released it already.

-Adam

[10bd01]

"Randomness" isn't "humanization". Proficient musicians don't "random", they may not play perfectly quantized but the way in which they don't quantize isn't "random", it's "human". It's unquantized according to how they express a specific music in a particular way, it's not random.

[adamj]

"Randomness" isn't "humanization". Proficient musicians don't "random", they may not play perfectly quantized but the way in which they don't quantize isn't "random", it's "human". It's unquantized according to how they express a specific music in a particular way, it's not random.

I completely agree. I added "humanization" to the description of my device a long time ago because I had seen similar features referred to the same way in other software products. The results of my device are "artificial" (not sure that's a good word either), but I still find it very useful in certain situations. Based on feedback, other people find it useful too.

Anyway, I removed the word "humanization" because, yeah, I can see it coming off as ignorant or over-hyping what this thing actually does. I was young and stupid

[CrystalWizard]

Seems that musicians do not purposely try to "randomize" when adding variation but on a very small scale "random" will make it more realistic. Probably some sort of complexity or chaos would be ideal. When we play a physical object every repetition will have some randomness inherently.
I like the "humanize" functions in Stylus RMX (especially) and in Battery 3. I generally add just enough to make it "interesting" more on a subconscious level than an obvious one.
I'll give your M4L device a try, thanks for putting it out there. Unfortunately i have to use the older version (ableton live ver 9 here).

[courtjestr]

"Randomness" isn't "humanization". Proficient musicians don't "random", they may not play perfectly quantized but the way in which they don't quantize isn't "random", it's "human". It's unquantized according to how they express a specific music in a particular way, it's not random.

Have to disagree with you here. Humanization does involve some randomization within bounds because no human can play perfectly quantized notes. The goal of humanization is to get rid of that robotic sound that comes from playing every note exactly on the beat. Ultimately, humanization has to do with adding the imperfections that result in a human playing a piece of music. What you are talking about is expression, sometimes playing just before the beat, sometimes delaying the note to give a certain emotional feel to the music. That too is in the realm of humans but I would not call that humanization. It goes beyond that.

[10bd01]

Have to disagree with you here. Humanization does involve some randomization within bounds because no human can play perfectly quantized notes. The goal of humanization is to get rid of that robotic sound that comes from playing every note exactly on the beat. Ultimately, humanization has to do with adding the imperfections that result in a human playing a piece of music. What you are talking about is expression, sometimes playing just before the beat, sometimes delaying the note to give a certain emotional feel to the music. That too is in the realm of humans but I would not call that humanization. It goes beyond that.

Randomness is robotic. It's algorhythmically generated - created by a robot - and it's not even truly random, let alone human.

I'm not talking about expression, I'm talking about the miniscule inconsistencies in quantization that are still not random. All of those inconsistencies are created through the intention of the player, and their intention is not to play random. It's not chaotic - that's what's being sold but it's not actually true. Even the imperfections are motivated/influenced imperfections, not random imperfections.

What you end up getting as a result of this train of thought is replacing a robotic sound that comes from playing every note exactly on the beat with a robotic sound of playing every note "randomly" off the beat. They're both robotic, you're just switching out a highly proficient robot with an error-prone robot. They're both artificial and both sound artificial, because they are.

There's no "humanization" going on here, I just wish they would call it "randomization" like the fellow above does.

IMO a "humanization" plugin shouldn't be robotically randomizing, it should be humanizing.

[courtjestr]

Have to disagree with you here. Humanization does involve some randomization within bounds because no human can play perfectly quantized notes. The goal of humanization is to get rid of that robotic sound that comes from playing every note exactly on the beat. Ultimately, humanization has to do with adding the imperfections that result in a human playing a piece of music. What you are talking about is expression, sometimes playing just before the beat, sometimes delaying the note to give a certain emotional feel to the music. That too is in the realm of humans but I would not call that humanization. It goes beyond that.

Randomness is robotic. It's algorhythmically generated - created by a robot - and it's not even truly random, let alone human.

I'm not talking about expression, I'm talking about the miniscule inconsistencies in quantization that are still not random. All of those inconsistencies are created through the intention of the player, and their intention is not to play random. It's not chaotic - that's what's being sold but it's not actually true. Even the imperfections are motivated/influenced imperfections, not random imperfections.

What you end up getting as a result of this train of thought is replacing a robotic sound that comes from playing every note exactly on the beat with a robotic sound of playing every note "randomly" off the beat. They're both robotic, you're just switching out a highly proficient robot with an error-prone robot. They're both artificial and both sound artificial, because they are.

There's no "humanization" going on here, I just wish they would call it "randomization" like the fellow above does.

IMO a "humanization" plugin shouldn't be robotically randomizing, it should be humanizing.

Any discussion I have seen about humanization and music involves some form of randomization. The inconsistencies are a result of humans not able to the keep perfect beat. Now you could argue that the inconsistencies are not perfectly random, but they are certainly not intentional. At least I have never seen any discussion that states that they are.

When I search for humanization and music on Google, every hit I find has some aspect of what I describe. Can you point me to a humanization plugin or algorithm that does what you describe? Or any discussion related to your point of view?

[10bd01]

Any discussion I have seen about humanization and music involves some form of randomization. The inconsistencies are a result of humans not able to the keep perfect beat. Now you could argue that the inconsistencies are not perfectly random, but they are certainly not intentional. At least I have never seen any discussion that states that they are.

When I search for humanization and music on Google, every hit I find has some aspect of what I describe. Can you point me to a humanization plugin or algorithm that does what you describe? Or any discussion related to your point of view?

I'm not arguing that the inconsistencies are intentional, but as you say that they're not really random, but also that part of why they're not random is because the motivating energy behind the mistakes isn't random but intentional. If I'm surfing and I make a mistake and I fall it doesn't happen "randomly", I'm not going to fall by spinning around three times, breaking both of my big toes, falling on my face and then falling off the surfboard. The mistake is informed by the intentional action that's happening when the mistake happens - it's not random, it's within constraints and is caused by actual reasons, which may be highly complex but are not random.

Same here - every discussion that involves humanization I've seen goes to randomness. And they're all wrong IMO. I think they're well-intended, but randomness is basically another "god" to explain what they don't understand, and a convenient and - frankly - lazy way to say you're "humanizing" something when you're - in fact - not. Assuming there's no pattern, consistency or form because you can't see or detect it doesn't mean it's random. I think the fact that the humanization discussion always appears to default to randomness is harming the *very small* industry of midi humanization because it stops people for searching for a true humanization by assuming they've already got it by using randomness.

So no, I can't point you to anything because I haven't found it myself. If I were to dive into this and - what I imagine may eventually happen - is that funneling a ton of midi data from live human performances of proficient musicians into an AI/Machine Learning library like tensorflow will yield a still-imperfect but very close (depending on the amount of data available) result and could be called properly "humanization" because it's using actual data from humans as the source and core of it's determinations - there may still be algorithms involved but the algorithms are serving that core of actual human data.

I think that would be the closest we could get with the technology I'm aware of right now. Would love to see it happen. I'd even love to work on it but there's only so many things one can work on.

There's a great way to bring in income with this in a similar way to how plugin developers now bring in money through collaborations - how about having a Sugarfoot preset where Moffett will play your midi data? That the plugin has analyzed hundreds of hours of Sugarfoot playing and has extrapolated this into his "playing style" with all of the expressive and nuanced inconsistencies in Sugarfoot's playing. These types of midi performance plugins - with general presets, sponsored presets and tweakable parameters - IMO are a market waiting to happen and where I'd like to see this go.

At the moment we have a bunch of robots tweaking out and randomly spitting out artificial non-random randomization and being packaged as "humanization". A total misrepresentation of what's actually going on.

[10bd01]

[DELETED] - Accidental double post.

[tonymax]

this one is subtle....

https://rastsound.com/downloads/naturaliser/

[courtjestr]

Any discussion I have seen about humanization and music involves some form of randomization. The inconsistencies are a result of humans not able to the keep perfect beat. Now you could argue that the inconsistencies are not perfectly random, but they are certainly not intentional. At least I have never seen any discussion that states that they are.

When I search for humanization and music on Google, every hit I find has some aspect of what I describe. Can you point me to a humanization plugin or algorithm that does what you describe? Or any discussion related to your point of view?

I'm not arguing that the inconsistencies are intentional, but as you say that they're not really random, but also that part of why they're not random is because the motivating energy behind the mistakes isn't random but intentional.

From the way I read your initial statements, it seemed to me that you were arguing this exact thing. You did state that "All of those inconsistencies are created through the intention of the player, and their intention is not to play random." I conflated that to say players are intentionally and consciously putting inconsistencies into their performance to derive a certain effect. I also inferred this based on your statement "It's unquantized according to how they express a specific music in a particular way". To me, the words "intention" and "express" imply a conscious effort. My argument was that those inconsistencies are not conscious nor are they are intentional but a result of many complex factors such as the distance from other performers, tiredness/alertness, even the distances between keys on a piano, or strings on a guitar.

I will agree with you that randomness is not a perfect substitute for true humanization. You would argue that it is no substitute for humanization. You are a purist and that is fine. For me, in many instances, it is a good enough approximation. For the time being, until we build those humanization AI machines, I am willing to accept the substitute. I still feel it is better than rigid on the beat quantization (unless that is the sound you are going for).