Hi,
I did the first minutes of Wagner’s famous Ouverture. It’s just the original score copied into Dorico, and NotePerformer. The only difference being the measured tremolos/repetition, that I have to write in expanded form, since Dorico is not yet able to correctly decipher the abbreviated form.
Not what I would consider a finished mockup, but impressive for the effort/result ratio.
Wagner - Ouverture from Tannhäuser (NP)
Paolo
A comparison with the VSL VI collection, therefore without the NP artificial intelligence overlapping Dorico’s humanization features:
Wagner - Ouverture from Tannhäuser (VI) (ROUGH DRAFT)
And with VSL BBO, an ensemble library with no solo instruments. The strings are, however, the new Synchron Strings Pro:
Wagner - Ouverture from Tannhäuser (BBO) (ROUGH DRAFT)
Paolo
Really interesting comparison, thanks for this! It would be interesting to hear this same Dorico file with other libraries, if you feel comfortable posting the file.
This is the first time that I have listened to an orchestral piece with the latest synthesis tools. (When I work in Dorico I turn off the sounds.)
I have to say that all three sound like the first reading of a very amateur orchestra which never takes breath. Particularly in the third example, it is clear that the players are not listening to each other, or even following a conductor!
I assume that these are just proof of concept examples. The string tone of the third one is promising, and the winds of the first generally preferable.
I shall be interested to hear refined versions.
David
NotePerformer is the only one that currently does some analysis and interpretation, as I understand it.
However, Dorico already provides settings to accent important beats of the bar and stress markings, and for the ‘overlap’ of legato playing. It also has settings for ‘humanized’ playing by varying the timings and dynamics.
So it will be interesting to see in the future whether it is to Dorico that we should look for more sophisticated interpretation, such as breathing and phrasing, or to the sample libraries themselves.
Yes, and this is why I specified “rough draft”. This is as Dorico + the selected sound generator produces it, with no further work on it. (I’m cheating: I did some mixing, to level things a bit, and many tempi, to make it more meaningful).
As an experiment, it may be interesting to see at which point the machine can help us. From there, it’s our job as humans to add what’s missing. Adjusting tempi, notes attack and release, varying articulations, making sustains breathe with a touch of messa di voce. Details, but unfortunately still a lot of work.
Still on Wagner — this is a mockup I did with Dorico and the VSL VI libraries; editing is at a more advanced stage:
Wagner, Tristan, Prelude to Act III (Dorico, VSL VI)
Paolo
This sounds much better, though the weakness of the double basses gives me the impression of D-flat major at the beginning.
A nice realistic touch is the poor ensemble between the first and second violins!
But in bar 11 we need to hear more of the first and third horns, and similarly the clarinet and oboe when it comes again are too weak.
There is still a sense of lack of direction in the playing.
How long did it take you to get to this stage?
David
There is something I really miss in these examples (and actually pretty much all mockups, regardless of the VSL), and it’s probably something that’d be very difficult to implement, and that’s the tuning, especially at the beginning of Tannhäuser. No wind section, or any group of instruments worth its salt, would play chordal passages like that without adjusting the pitches to produce perfect intervals, especially the thirds. With most computer-produced mockups I’ve heard, we’re stuck with good old equal temperament, with its ugly thirds unworthy of real musicians. I really hope that people using these VSLs don’t get so trapped in this bubble that they come to accept and even prefer what comes down to poor ensemble intonation, which is solely based on the tuning of an instrument like the piano which has the limitation of having to have its pitches set in stone.
I was actually relieved to read another post just after this one called ‘Winds detuning’ which discusses this very point and brings up Hermode tuning. I’m glad that there are people our there who are working with it. I used to experiment with this in the Aria player but, as was remarked there, you’re limited to a single tuning system and a single reference base pitch. I’ve heard Hermode tuning applied to piano music, which is really bizarre: the pitches change according to the context!
David,
Thank you for pointing out at some of the weaknesses of the Tristan mockup! I’ll try to fix these points, and other similar to these ones.
May I ask you to elaborate on this? Do you mean they don’t blend well enough? I have to admit that I’ve tried to find a separation between 1st and 2nd violins, being advantaged, in this, but the default VSL setting of each section being positioned at the opposite sides of the stage.
The Tristan mockup was something that I did in my spare time for about two weeks, and then some other touches later. Before that there was the note entry phase, but I don’t remember how long it lasted.
As for the weakness in the strings attacks: I have been accustomed to more robust attacks in this piece. Then, a performance led by Robertson convinced me that this lack of contour can work better. And I went for sort of Debussesque freedom in time, with the goal to achieve a general sense of softness. It’s the only thing in this mockup I am, at the moment, convinced about.
Paolo
Hey, that’s me who started that mess! 
Yes, I’m aware of this issue. I’m already working to edit some of my mockups to see if it can be achieved easily. Otherwise, it is really a lot of work, blindly done in the pianoroll page.
Paolo
A bit off topic but maybe interesting anyway – there are actually pipe organs with the hermode tuning system incorporated. A small motor tunes each pipe in real time during the performance. If you scroll down a bit on this page
That would be very useful to play Baroque ensemble music with a piano and an electric guitar!
Paolo
The second violins are noticeably behind the firsts in starting their notes. (I like the seconds on the right and have positioned them like that in my orchestras for nearly 50 years – Klemperer convinced me, and Mahler and others did the same!)
In terms of time taken, I wondered how many hours one has to spend, after inputting the notes, etc, to make the results sound musical. This is largely why I have never bothered to use synthesis in Sibelius and Dorico for anything beyond helping to discover the wrong notes!
David
David, thank you for advising about the violins timing!
It takes a lot of time. I would say that if one doesn’t need a sonic rendering of a score, because the piece will be performed by real players, it is too much time, better spent after something else.
If the score will probably remain in its digital form – either because with no chance to be performed, or because the final media product will not require an actual orchestra – working on it becomes unavoidable.
There is also a third case, and it is submitting scores for competitions. While it is supposed that the members of the jury will read the score and hear it with their inner ear, since they are always asking for a mockup they will probably do a first selection by listening to the mockup. So, the better it is, the more chances one has to be actually considered.
Paolo
There’s a rather large digital organ maker who has hermode tuning available as well. There’s a video floating around of someone demonstrating various chords and demonstrating how the computer adjusts the intonation essentially in real time. It’s fascinating. Touted as a major advantage for accompanying choirs.
Having something like this as a continuo instrument would completely change how one plays continuo. A good player has to take the tuning of the instrument, the tonality of the piece, the chords being played and the other instruments all into account. Hermode tuning would take away a lot of the skill and understanding required to do this well, although it might very well create new problems (which I’d enjoy tackling…).
The problem is: what does the system take as the reference in deciding what to change from chord to chord? Also, how does it get back to the original pitch for the ending?
It reminds me of a recording of a Lassus unaccompanied choral work I was once involved in as the producer. The work was about 15-20 minutes long and highly chromatic. It modulates a lot and the basic pitch drifted either up or down during the takes. This was not always apparent to us during the recording sessions, but editing the takes together was a nightmare. Having been a member of a choir that sang simpler unaccompanied music than this, I came to the conclusion either that Lassus’ choir must have been hypertalented to stop this wandering of pitch, or the effect was par for the course!
David
This is why I continue to hold out for more sophisticated play engines that can render notation in a realistic way. Whether that’s to be found in NotePerformer or in Dorico itself done the road, there’s absolutely a massive market for increasingly realistic renderings with zero extra work.
The second scenario you describe is still the exclusive realm of a DAW at the moment. Perhaps it always will be.
I’m all for improved computer rendering, but when one considers the effort that shaping a live performance almost inevitably requires, one wonders whether AI will ever conquer the “beauty is in the eye/ear of the beholder” problem completely.