I just wanted to inquire about the best workflow with levels / compression / clipping, etc. I know the NP4 mixer optimizes levels across libraries, but does it account at all for clipping or do any sort of compression to leave headroom?
I’m asking because I own BBCSO Pro and the NP playback engine for it and I was going to post @dan_kreider’s sample file from this thread for comparison. Using that file with NP4 with the BBCSO Pro engine loaded, all Dorico levels flat, NPs reverb, and no other plug-ins in Dorico’s mixer (so the default reverb and compression removed) I end up with this:
I suppose this result isn’t at all unexpected with a larger ensemble like this, but it does mean NPPE isn’t quite as simple as load the library or libraries and go. All the individual levels in the NP4 mixer seem ok, it’s just the cumulative output that is pretty clipped.
Obviously I can drag the Master fader down or add a compressor, but I just wanted to see if NP4 is doing any of that internally as well, since I know it is doing some of its own balancing.
(It sure would be nice if Dorico’s mixer had dB levels and the usual green/yellow/red indications.)
although I can’t run the full Cinematic Studio orchestra, I thought I’d try the solo strings on their own and there are some good things here, particularly the very natural attack at the beginning of legatos for instance and intelligent phrasing. However, it seems portamento has not been implemented (should be activated with velocity on under 20 for the 2nd note normally but I find no way to activate it) which is a major drawback with this particular library if true. Perhaps Arne would be able to comment at some point. The same question might also be asked of the BBC SO plug-in – I didn’t test that function there yet.
Incidentally, the ambient hiss level seems rather high with CSSS as well – much higher than when run natively through Kontakt.
Fred, what is the zoom level in your audio software?
I know that in studio one you can change the amplitudinal-zoom (for wont of a better term) so that things can look like they are clipping when they aren’t.
Sorry, I didn’t save the file to go back and check. It definitely was clipping though. Even in the Dorico mixer you can see the “Lead” fader where NP is returning the signal has completely peaked. As I had no compression and the Master fader in Dorico was zeroed, it looked the same.
When I opened the BBCSO Core MP3 Dan posted in the other thread along with my file, there was an obvious difference.
He said he didn’t create that file (bottom green one) but I’m guessing it had the default Dorico compression vs my blue one with the levels at 0 and no compression.
I think that’s because BBCSO is much more bottom-heavy than NP so the loudness produces a greater amplitude. This could be room for improvement in our software.
I’m wondering if this is BBCSO-Core or NP4 that is doing this (I have no other software than Dorico, so cannot test in a DAW), but I have a passage of 8th notes, under measure-long slurs.
At the end of each slur, the sound very suddenly drops down and wants to cut.
It’s as if the violinists were stopping their bows at the end of each slur and having to start up again for the next slur.
Is there a way to fix this? or is it an inevitable artifact of BBCSO’s string samples?
I would be curious to hear if anyone has a solution to this. I’ve found the same thing in EW HOOPUS. The sounds are way too agressive at shortening things that aren’t slurred. Normally, string and wind players still play very connected even if not slurred, since slurs for these instruments represent very specific instrument techniques, as opposed to on piano where the the slur/non-slur distinction is a little fuzzier and subject to interpretation. I wish NP would not be so quick to switch to the shorter articulations across the board, and if there’s a way to control this that I’m not aware of, I’d love to spend some time dialing in my preferences.
If one loads the BBCSO violins directly in Dorico using the legato patch, the problem is much less pronouced? Also with the native NP sounds there is no such problem. Seems to me like NPPE is overdoing the phrasing…
this is what I was saying a few days ago. However, the problem can be reduced or even eliminated simply by lengthening the notes in the Key Editor, at any rate with the BBC SO where I’ve been trying it. It’s one step less far than putting a tenuto marking or something. Of course the exact behaviour will depend on what choice of shorter articulations the library offers. I’ve found that purely in that respect, even with the current limitations of non-programmable boundaries, it’s possible to get a somewhat higher rate of the correct shorter articulation in a well-programmed map where there is a wide choice (in libraries like VSL) than NotePerfomer necessarily managed but I’m not underestimating the overall capabilities of these engines - I have been more impressed than I’d expected to be honest so far with my natural scepticism
actually I wouldn’t make so much of a distinction between a DAW and Dorico which has the majority of DAW functions you need. I’d simply say remove 90% of the problems in using the libraries full stop. As libraries are in the first instance designed around playing live, any sort of recording for later MIDI manipulation shows up the bugs and specific difficulties of the individual ones whatever you use. What is new is NPPE’s attempt to intelligently analyse how the music should be performed by using a wide range of information gathered in the read-ahead and in certain areas with the libraries I’ve tried, the achievement is impressive. As some posts have demonstrated, things are not quite perfect yet but I’m pretty happy with what’s been done so far.
Thank you very much for your time and the detailed reply on my comment!
Well, obviously the music World heading to a software of new generation DAW-like notation applications like Dorico, MuseScore 4, Overture, the upcoming Encore 6 (those two most probably will fade away from the market), Staffpad… If programs like Sibelius and Finale don’t move forward they will be at risk of loosing their entire market share. This is the future.
The increasing Dorico user base, and the DAW features in MuseScore 4, show that the notation software isn’t a “Microsoft Word for music” anymore… Yes, there are people who are using those apps just as MS Word, in some cases me, too.
Won’t be good for NP and NPPE if you focus your attention only on the “MS Word” side of the Notation DAWs. That way you may put at risk the future of your product.
Currently you are the one to offer such multi-platform AI solution, but if are neglecting the DAW side, sooner or later another company will come up with a product that covers both aspects of the Notation DAWs. Once such application becomes available, NP will become unnecessary…
I know that your company is a very small, but you could always get into collaboration with companies like Steinberg and the team behind their VST products. Very often different companies collaborate in order to create something innovative and successful. If you can’t afford to hire more developers, the only solution is a collaboration with a bigger company. I suppose Steinberg and Yamaha wouldn’t mind to collaborate with on a new AI player that will support the whole potential of Dorico and possibly Cubase/Nuendo. Why not?!
Well my point about control was oriented more about dynamic curve - we to decide how quiet is ppp, or how loud is fff… of course the change of the dynamic curve will influence the timber, too…
an attack aspect. Actually some control over these two aspects will provide good amount of personal fingerprint on the playback.
Here’s what I’m getting on my system using the Dorico file Dan posted here. For all audio exports, Dorico’s levels are flat. (In preferences.xml, the defaultPluginOutputGain is set to 1) From top to bottom:
NP4 with Dorico’s default compression
NP4 with no compression
NPPE with BBCSO Pro with Dorico’s default compression
NPPE with BBCSO Pro with no compression
Spitfire’s BBCSO Pro Playback Template with Dorico’s default compression
Spitfire’s BBCSO Pro Playback Template with no compression (The Spitfire files are missing Piano and Vocal as well)
#6, the bottom BBCSO Pro file using Spitfire’s Playback Template (not NPs) without compression sounds terrible as it’s very distorted. Here’s the ending:
Even the regular NP4 audio without compression exhibits a bit of clipping on this file though. The level issues are obviously correctable as long as the user is aware, but are the NP4 and the BBCSO Pro NP playback engine designed to be paired with Dorico’s default compression and levels? I think it’s fairly safe to assume the Dorico playback template developed by Spitfire is calibrated for at least Dorico’s compression as it sounds terrible without it. By the looks of what I’m seeing, I’m assuming NP4 and NPPE aren’t actually designed for flat levels and no compression. (Or at least the BBCSO Pro playback engine.) Is it assumed the user will naturally set master levels and compression as needed? Or are they supposed to be “set and forget” with the default levels and compression settings? Just curious if there are intended values that the libraries are calibrated for.
this is only the 1st and 2nd violins and violas, from BBCSO-Core.
other than this one artifact, it plays back perfectly, without hiccups or any issues. (the piece is for solo violin (NP3) and string orchestra (instruments from BBCSO-Core).
Even looking at the waveform on Dropbox as it plays, you can see it’s suddenly dropping off on the last 8th note of every measure. It’s almost as if it isn’t playing that final note.
With just the strings from NP3 it plays smoothly.
EDIT:
I tried it with another passage, in a large orchestral work, with the same type of texture, but in woodwinds, and there didn’t seem to be any problem with the playback there (though the oboe was a TAD scratchy… it was the climax of the piece and the dynamics were VERY loud, though notated dynamics, not adjusted in the Play window)
It may be an NP4 issue. I used NPPE with my Synchron Prime, and in legato eighth-note passages, they are too short, making the passage sound choppy. I did not notice this artifact when using VSL’s Synchron Prime Dorico playback template
Wallander probably knows this, but others might not.
I’m not sure how NP deal with it…but for everything else…
Legato has always been weird in Dorico if the library uses a keyswitch/footswitch/channel bounce via expression map.
I.E. A passage of four quarter notes under a slur.
Dorico plays “la la la ta”, where it should be “ta la la la”.
Unless you script up an instrument to deal with it specifically for Dorico, the trigger starts too soon for a phrase, then releases too soon.
Not sure how NP deals with slurs, but over the years I’ve learned to simply strip out any automatic playback settings in expression maps for slur/legato. In Halion I can add my own delay script to hold up the legato triggering event a few ticks. Another work around is to host stuff that can’t be user scripted in something like Bidule or Kushview and delay the trigger there.
If I want scores for an ‘all stock’ Dorico installment…I just impelement legato the old-school General MIDI way, and let Dorico add a little length to such notes for that ‘overlapping/crossfade’ effect between notes. Some libraries hate it (cut out or sound bad) if the same note is repeated and there is overlap…but for 99% of the stuff I’ve ever tired/used…old school General MIDI MIDI legato (overlapping notes a little) is alrite.
One of the first things I do after a fresh Dorico install is go into the HSO String expression maps and DELETE that legato node! I’ll then go into general playback settings and instruct dorico to add a few ms to the length of notes marked legato. Out of the box…the setup wrecks phrasing pretty bad in my opinion (anything with strings and the default slur or slur-style phrasing marks, the ‘legato’ key switch is a different bowing speed, and the way it’s implemented is poor). Take the legato node out, and scores involving strings already sound 95% better.
There may be some natural volume discrepancy between NP and NPPE, but what users have shown is too much, as is the case here. I need to look at adding another layer of correction. There shouldn’t be this much difference in volume.
It’s natural for the Spitfire template to have a different level because it’s unrelated.
Unfortunately, it’s impossible with a set-and-forget setting for output levels. It’s not ensemble-agnostic, but a full orchestra is much louder than a chamber group. In a production environment, you would normally export the sound with a decent amount of headroom (no overshooting) and use a wave editor with tools (compressors and limiters) to bring the volume to the desired comfort level. Those tools will duck the signal in case it overshoots momentarily. If you only normalize the wav file with no compression or limiting, it will be much lower in volume than a CD or a mp3 from iTunes.
With that said, we’ve taken steps to make NotePerformer a decent starting point, so you can often export it directly. But it seems that NPPE is a bit louder, which is unintentional.
I don’t doubt the market will change in various ways. It’s unavoidable and nothing I will lose sleep over. It’s just the way of life, and who knows what role software like our’s will play in one-two decade. Before solving those problems, you can type words into a browser A.I. which generates a symphonic work to your instructions, or upload a crude song and get a cover in any possible music style.
I don’t have any ambitions for a big team or going into large-scale collaborations. I’m happily running a small business where I can make the rules and do interesting things. I enjoy the creative process but not the attention it brings, so I like it this way. Besides, big teams have creative limitations, and it’s well-established that they’re better suited for recycling others’ ideas than thinking outside the box. It’s not as simple as adding two teams and having the best of both worlds, but it can also be the opposite.
Sample libraries don’t have freeform dynamic curves or timbre curves. They have 2-4 layers of samples and volume scaling. To some extent, you’re the victim of limitations in NPPE that we may need to improve, to another extent you’re the victims of limitations in a sample library. If you want that level of control, our built-in sounds are still there.
Regarding dynamic range, we don’t present such features because we paint ourselves into a corner if users can access features we want to control internally and have a chance of improving or updating. If you want more control over dynamics, there’s a lane for that in Play view where you can draw exactly the dynamics you want.
