Interesting thread for sure. I think there are a few different ways to look at this and what is trying to be achieved. I have nothing much to offer from a technical advice standpoint but the imperfections are something I think about a lot.
To me, human imperfection can be modelled with random or algorithmic manipulation of some sterile, for lack of a better word, source data. I don’t personally think that is all that useful though. It can at best represent a poor human performance. Haphazard hesitations or dynamic changes through a phrase are what happens when a performers attention is focused on the wrong thing. Perhaps just learning a piece or just generally distracted. Or perhaps they just got too caught up in trying not to make “mistakes” or are rushing (ahem, me) or don’t actually have sufficient motivation or connection to approach it musically. There is also the theoretical element of understanding what is actually going on from a musical perspective and where your focus should be directed.
When there is intention behind the phrasing, pauses, dynamic development, balancing of voices, etc. basically the musical expression of performance, that is when you get something really worth putting the effort in to achieve. That is much harder however as it cannot be achieved randomly or (traditionally) algorithmically because it requires intent. The decisions must be guided. Actually, the key point is they have to be decisions, not accidents. This is of course perfectly possible to be done in an offline manner by editing midi data or to a lesser extent based on interpretation of score annotations. However, the current best interface to achieve this is in my opinion real-time input from an instrument (acoustic or midi or otherwise) by a practised performer.
Performers develop an active connection with the sound produced by their performance. Their decisions are based on instant auditory feedback on the impact of their actions. Tiny intuitive adjustments are made in a subconscious way based on the skills developed through intentional practice over a period of time. Music is, after all, a listening experience, we can’t achieve the same level of control and nuance when it is approached visually (in the case of editors) or not in real-time (in the case of looping or assessing the results by playing back modifications). There is too great of a disconnect between the actions or decisions and the resulting sound.
I guess ultimately that is the difference (as someone mentioned up thread) between human imperfection and human perfection in musical performance. There will always be a level of imperfection of course but that is not actually the thing we are (perhaps, should be) trying to add with the humanise button. The flaws may make it sound human but they don’t make it sound “musical”.
The crux of all that for me is the fact that emulation of what can be achieved better and more easily through existing means is perhaps a misguided goal. If the goal is to bring “human” nuanced expression to the phrasing of melodic lines, the solution is to use a real-time expressive tool to do so. On the other hand, if the goal is to break free of the grid and bring greater dynamic and rhythmic variation to our music, there are loads of cool and interesting techniques and tools, which other people are much better versed in than I, to do that. It’s probably not going to sound like Satie and that’s probably a good thing.
To make it clear, I place no greater value or validity on real-time performed music etc. Generative, random, locked to a grid, noise, a hoover through a distortion pedal, whatever. It’s all equally intriguing and valuable to me. The fact that I am writing this while procrastinating on finishing some recording I can’t get right means that currently, perhaps I prefer the concept of generative or sequenced music.
