Thanks for sharing that - it’s consistent with other renders I’ve heard from NP5. The strings are not good at all, but the other instruments are better than NP4.5 (IMO!).
the strings are undoubtedly the main problem area (as they are with native NP despite some improvements this time round) being often rather acidic but the degree to which that is an issue depends on the context and on the whole, I can live with it in this work.
1 Like
I agree that NP5 has some promising qualities, and the clarity it brings in certain passages is definitely encouraging. Regarding the Cinematic Studio Series Strings (solo and section), I do think the dynamic matching between shorts and longs is noticeably improved - it feels more cohesive and musical overall. However, the expressive legato, which is one of the standout features of this library, feels like a big step backwards in NP5. It’s a shame, because that was a major part of what made the library so compelling. Hopefully this is something that can be addressed in future updates, as there’s a lot of potential here if that expressiveness can be fully captured once again.
4 Likes
What Arne has done with NP5 is great. I can now use my BBCSO Core and Spitfire SSO 2024 in full orchestral works on my PC-an impossibility previously with my PC; since the 2 of them cost @$1,000, this is welcome news indeed. I could only use individual sections or instruments. Also, the new instruments added to NP5 are nice. And, while there is some difference in the quality of the as compared to the full versions, with a bit of ‘tweaking’, they can sound quite good. Also, if I want more robust ‘real’ versions of my VST libraries for playback, I have the Dorico templates-BBCSO Core, Discover, Berlin Free Orchestra & GPO5 (thanks @benwiggy!), my VSL templates created with the Dorico Wizard and Kontakt. I have more options than ever before in a notation program. So, THANKS Dorico team and Arne!
1 Like
Yeah, @alexsMusicComps as @ikos said, NotePerformer is not really AI by the definitions we would use today. It was described as AI shaping before things like Midjourney and ChatGPT existed, before we developed our modern definition of AI based on machine learning and what those types of platforms do. The word AI used to be used a lot more loosely than it is today, encompassing a much broader series of situations.
NotePerformer works as a rule based system like Dorico expression maps, only with many more rules, more comprehensive. It is given a bunch of rules “if X happens in the music, then do Y”, and those rules are constructed by a human being to make sure they usually have a positive effect rather than a negative one.
1 Like
Ok and if this would be the case, why we do have a preroll like in all AI training Engines today?Has nothing changed in NP5? I somehow cannot believe that, its smells alot like more than expression maps.
For everyone who likes to the see the difference between the default basic Reverb and a good Reverb Simulation i made a short 12 minute tutorial and demonstration how much it truly improves by deactivating the default reverb and adding a realistic one:
3 Likes
I wouldn’t say there are any direct “competitors” in the notation space, unless you switch notation programs entirely, but my experience is that libraries can work extremely well run direct by Dorico as long as they are:
- Consistent - common set of articulations, ideally pretty comprehensive, and respond with consistent delays ideally
- Agile - are able to play fast passages without falling apart (ex. written out runs or “flurries” of various kinds)
- Balanced - articulations balance correctly within each instrument (ex. col legno should not be as loud as snap pizz), instruments balance properly with each other (harp should not be as loud as trumpet or trombone at the max), instruments balance properly within their range (ex. low flute should not be as loud as high flute)
So if the virtual instruments meet this criteria, they can work just fine being driven by Dorico. The problem is that very few instruments meet this criteria. One of the few vendors that do well on all three fronts is VSL (Vienna Symphonic Library) which makes expression maps for their products for Dorico as well.
I mostly use VSL libraries in Dorico, driven directly by Dorico Expression Maps, and I’m able to get good balanced playback out of the box. Missing some human shaping, but good - and I can add the human shaping myself. I have never really used NPPE at all aside from some testing, because I never had a need for it with this solution.
If you try using libraries that are not consistent, agile and balanced, driven directly by Dorico, the results will be quite disappointing (and this goes for the majority of popular libraries out there), because so many things will be off if it is overwhelming and you don’t know where to start. If you use libraries that are consistent, agile, and balanced, the out of the box result will already be good, and it just needs some extra manual musical shaping to push it into the “great” zone.
4 Likes
Because the rules can do a better job when they can be defined to act on future events and not just what is happening at that moment. You can do things like “if after this note ends, there is a rest followed by an accented note, do Y” (where Y might be a slightly different shaping knowing what comes next). With no preroll, the rules can only be written to act on what is happening now or what has already happened, not what is coming next in the music. Even basic rules sometimes might need to make exceptions based on what comes next in the music, and without the preroll they have no information about what comes next.
3 Likes
Ok accepted, but why is Noteperformer 5 after training working and replaying the signal even, when i disconnect my BBC Pro Library SSD? There is zero streaming happening or requirement to even have a BBC Pro library Harddrive of the original samples after Training or Preroll. Which is a sign of either a good training data set or a ultra large compression factor on the sample data. ? Sorry for my questions but i want to understand that.
NP5 pre-mines the samples into its own preload file. It does a subset, and stores them with slightly lossy audio compression. This is why they are so much smaller.
3 Likes
i would not say lossy. because arne said the compression isnt the problem, also double buffer does not change the resulting wave on my screens yesterday, but it removes round robbins and other microphones.
1 Like
It is not “training”, it is resampling essentially. Resampling is a technique where you convert samples to another sampler format by playing them on the original device and recording the output, and then mapping it to another sample player. IK Multimedia does this with their “Syntronik” product, where they played many notes on the original synthesizers and recorded them, and mapped them into their sampler so that when you play the key you hear the recording of the original synthesizer even though you may not own that hardware synthesizer at all. That is resampling.
So what NotePerformer is doing is playing notes to whatever VST instrument you are using (could be Kontakt or something else) and recording the output, playing every note to resample the entire instrument and basically creating a version of the library for “NotePerformer Sampler” instead of “Kontakt”. Then it uses “NotePerformer Sampler” to play back the library instead.
Resampling like this would normally violate EULA which is why they have the restrictions - it is encrypted and does some tests with the original instrument so that you can’t use it on a system that doesn’t have that installed. That way you can’t use this resampled version to get a “second license” for a library for a computer it is not licensed on.
4 Likes
More importantly, you don’t get the musical equivalent of missing or extra fingers in the resulting output.
3 Likes
Ah… interesting now all makes sense.. thanks for those information!
I hope my Tutorial video above will help to get the magic back in NP5 … for some people. I truly love the update.. Removing all Reverb Samples from the original samples to get a ultra dry signal is a brilliant move. I truly hate that the original Sample has all the reverbs in it..which is horrible when you reposition the signal in a spatial reverb simulator.
1 Like
I wouldn’t call this a positive in general. There is a good reason why ambient libraries and not dry libraries dominate the market - simulating a room around a bone-dry instrument is never going to be 100% successful. Some people feel differently but they are in a minority. Taking the room out of the libraries is actually bad from a perspective of realism - putting a fake room back is just not going to sound as good. Sure, dry samples do have advantages in terms of being able to position them wherever you want, but if you’re just going to put it in its normal orchestral seating position anyway, this isn’t a problem. You can also construct pretty good fake rooms for individual dry libraries to mix them with ambient ones. But I’ve rarely heard a mockup made entirely of bone-dry libraries, relying entirely on MIR or other solutions to place them, that has convinced me. Yet I use MIR all the time to add a second VSL harp (one of the dry Studio ones) when I need another harp on top of my Synchron harp. But that’s just part of the texture, nobody is going to say “that second harp sounds fake” when you can hardly hear it. The simulation is good enough. But when everything is made up of dry samples positioned in a fake room, there is something off about it that is hard to put your finger on.
I think it isn’t as much a decision for “improving” the libraries as it is a pragmatic one, to try to equalize out the differences and make dry libraries and wet ones as similar as possible so that they behave in a consistent way, as consistency is important for easy switching back and forth. But this pragmatic decision is unlikely to deliver a more realistic or better final sound - it just makes the libraries easier to work with by making them more the same.
5 Likes
@alexsMusicComps I’ll refer you to a video of Anne Kathrin Dern (a rather well known film composer) talking about the problems of making a library dry and simulating a room:
5 Likes
What exactly does the number of plugin instances in preload specify? If I select 1 plugin instance, does that mean I can only load 1 instance of an entire library, or is that a limit on the number of slots for each instrument?
Thanks!
I don’t know what technology NP5 uses for de-reverb, but I’ve tried out a couple of de-reverbing tools while mixing and they definitely affect the sound. It doesn’t come for free, in my experience.
2 Likes
She’s Great. So willing to share her knowledge and experience 
2 Likes