drastic eq changes but a great improvement.

Amazing, glad you're seeing a big improvement .

By the way, not sure everyone's aware but apparently Kawai have announced a successor model to the NV10s, called the NV12, whose headline feature is.... a soundboard! That's an exciting development that would probably have saved me a few modifications. Impeccable timing on their part, given that I only just got the NV10s, but what can you do .

Anyway, if I use the benefits of my current set-up as a baseline to what the NV12 might sound like in combination with pianoteq, I'd recommend keeping an eye out for it... in theory, it could be really great

Pianoteq always sounds great through my relatively inexpensive headphones but I wasn't happy with any piano VST through my Adam's T7V's.  Something just didn't sound right. I tried to manually change the EQ and could get something acceptable but still wasn't happy.  My home office is also my practice space.  Untreated walls covered in glass writing boards for my job.  Not the ideal room but it's what I have to work with.

Inspired by your post, I ended up borrowing my neighbor's UMIK-1 and went through a similar process.  I only did a single measurement and used the Harman curve.  Like you, drastic eq changes but a great improvement.  I plan on repeating the process with several measurements and I'll also try the DrToole target curve when I have a chance.

Thanks for the write up!

I remember a selling point of the NV5 versus the NV10 was better sound, largely due to the built-in soundboard.

I can believe that; one thing I learned in this process is that sound dispersion is more important to realism than sheer sound quality (at least to me), and a soundboard will be very good at that. The other thing the NV5 has going for it is that you don't have a direct sightline to the speakers, which avoids that "speaker point source" effect I was battling. Luckily it was easy enough to remedy these gripes on the NV10s... still, it's good that your NV5 doesn't have them in the first place.

I'm not going to mess with the actual speakers though. There's probably no need, on the NV5. Back when I was shopping, I remember a selling point of the NV5 versus the NV10 was better sound, largely due to the built-in soundboard. People often complained the sound of the NV10 left something to be desired. It sounds like you've completely remedied that.

can you share more technical details

Hey, sure! I'm happy that this is of interest, let me get a bit more into the practical detail

Room set up and EQ:

This is a room 5m*10m room with soft furnishings, the piano is close to one of the corners. I know from setting up my speaker systems that there aren't particularly crazy acoustics going on, so you suspicion that most of the correction would be caused by speaker internals is likely on point. Reading this post and confirming that the NV10s has the same components as the NV10, I can see that the cross-over routing is the weakest link in the chain. I used Kushview Elements as my VST host, and linked the left and right channels into separate 16-band AUNBandEQ instances (the PEQ that's installed by default on Mac). Here are the parameters I ended up with:

https://imgur.com/a/ioydlyu

Even though there's pretty deep cuts there, I noticed that even small changes to the parameters would make the sound artificial. For it to sound like an acoustic grand, the integration into the room acoustics had to be pretty seamless. So my guess is that you won't get good results if you copy my values - you'l probably have to measure your own room correction. Briefly, here's how I did that:

I used a calibration mic (UMIK-1, highly recommended), which I centered exactly between the speakers, on my piano bench, at ear height. I recorded a couple of measurement sweeps in Room EQ Wizard (learning that app is its own can of worms, but this video helped me get started: https://www.youtube.com/watch?v=HYMQ6M-Z5rM). I also took additional measurements slightly to the left and right to create vector averages so that I could get good results when I move around as I play. Room EQ Wizard can automatically fit EQ parameters to match your signal to a target curve, and I found that using the DrToole target curve (which is almost flat) gave me a good compromise of sounding "good" and "real". A more obvious choice like a harman curve sounded a bit too processed and cinematic to me. Again, I was surprised how even miniscule changes would have a big impact on the results, especially considering the amplitude of the corrections. I experimented and got the best results by measuring and correcting the left and right channel separately.

I didn't apply any room treatment, because after the eq correction, any remaining resonances felt natural, and similar to what you would get from an acoustic grand.

I tried to manually tweak the eq parameters here and there but found that while I could make notes sound "better", they lost a bit of their "natural" feel, so I ended up sticking with the automatically matched parameters.

Signal chain gear

After a bit of research I went with the Topping D50 III DAC, which has a reputation for extremely neutral and close-to-reference processing. In hifi-terms, it's modestly priced, but honestly for the purpose it's pretty overspeccd. A Topping E30 II would have been more than enough. Cable-wise, it was all about finding a short, well shielded cable; in my case, this one did a fine job. The signal chain is not perfect - when you put the volume to maximum, you can clearly hear white noise from interference being picked up - but when you adjust the volume to a level that won't take your ears off, not only can you not hear the noise directly, it also doesn't colour the tails of decays in any perceivable way.

It annoyed me that I needed to produce a high quality analog signal just to have it redigitised by a mid-tier ADC (96khz/24bit according to my research) so I spent a bit of time seeing if I could feed the line signal from my DAC straight to the speaker amplifiers. I think theoretically there is a way, but I would also have to create a combined and filtered mono signal for the woofer, so I left that rabbit hole for another day. I bet it would improve the signal quality a bit though.

The trick was to feed as "hot" a signal as possible - ie make the incoming volume as loud as I can without running into clipping. This keeps the signal strong relative to the noise levels.

Pianoteq changes

My go-to velocity curve is as follows (based on the memory of playing acoustics; I hope I'll have the chance to fine tune this when I can compare instruments side by side):

Global Velocity = [0, 12, 60, 80, 100, 112, 127; 0, 3, 32, 64, 96, 122, 127]

Apart from that, one thing that really helped me fit the sound into the room was to match the reverb to the natural acoustics of my room ("Dry room" preset with duration bumped to 30ms, and mix at 0dB). I experimented with recording an impulse response of my room and load that into the reverb effect, but this ended up having too much distortion to be workable; but in principle it seemed to head in the right direction, so maybe I'll revisit this.

Finally as a point of personal preference, I disabled the phase flip in the Delay effect - my ears register these phase flips as a telltale of digital processing and so I found the illusion of an acoustic grand more convincing without it.

Speaker modding

Oh, I had to bring out some advanced audiophile gear for this one... just kidding, I covered the tweeters with two heavy books, and redid the room correction afterwards to make sure the muffling is not too extreme. The main effect of this was that it spread the perceived source of higher-frequency sounds around enough to trick my ears into thinking it comes from a larger, diffuse, resonating body, as opposed to a point source.

Coupling the piano to the wall was a similarly advanced technique: I wedged an old wooden chopping board between the back of the piano and the wall. I felt silly doing it and I feel silly writing it now, but it really worked. I experimented with different positions and found that the center of the piano, which happens to align with a fixed point of the drywall, works best.

By the way, obviously a section of drywall doesn't have the same colouration as a piece of fine tonewood, as a soundboard would be made of. But that's fine - the colouration has already been done by the pianoteq model. We don't want additional colouring, we just want effective dispersion - and it turns out that even humble materials can be quite good at this. See this video for another rabbit hole: https://www.youtube.com/watch?v=zdkyGDqU7xA

Hey all, I promised an update once I had set up Pianoteq with the NV10S.

Wow, thanks. For people with a similar instrument (or intention to purchase one -- neither or which is me), can you share more technical details. For example:

- room dimensions, floor kind and furniture (relevant for next)
- software used for EQ and values (frequency and attenuation values) -- I know this will need to be different in different rooms but given your "aggressive" configuration, I suspect that the majority comes from the NV10S speaker response and only a small part from the room setting
- selected DAC and cables
- velocity profile (if changed) and other pianoteq settings if different from default
- how you muffled the twitters and how you coupled the woofers
- anything else you would need to tell to somebody trying to replicate what you have done

Thanks again, excellent work!

Q: How does the lack of key-off velocity affect things? To test that, I wrote a script that would cause the release velocity to match the preceding attack velocity… and I didn’t really like the result. It did, as expected, modify damper effectiveness so that the keys would be silenced more slowly, but I couldn’t make this correspond with any playing intent. I decided I can live without it

Q: How does the built-in sound compare to Pianoteq’s Kawai model? The built-in instruments are outstandingly playable sample libraries with some harmonic resonance modelling built on top. If kawai were to release this as a VST, people would rave about it as much as they rave about the Garritan. Part of this playability is achieved by flattening the samples: attacks are cleaner, decays quicker, resonances tamer than they would be on a real instrument, in part to more clearly translate the playing intent to sound. So overall, very very good job with these sounds. But to cut a long story short, Pianoteq is way better. It manages to maintain the complexities of an acoustic grand without muddying the player intent, and sounds a lot more organic as a consequence. You can’t go back to the built-in sounds after this. The main differences is in long sustained chords, which bloom much more convincingly in Pianoteq, and very fast, technical passages where the “secondary” resonances begin to outweigh the direct sounds: in a sample library, this starts to sound like “black midi” - like soundfiles that get clipped to early - and in pianoteq, it sounds like a continuous recording of multiple notes, because striking a note for a second time doesn’t stop and restart a sound: it modifies an ongoing sound. So at the risk of coming to an overly safe conclusion, it’s a slam dunk for Pianoteq.

Q: How close to an acoustic grand is the sound? So here we get philosophical: Should it sound like I’m playing a concert grand in a concert hall, or like I’m driving a well mastered piano recording, or like I put a concert grand in my way-too-small living room, where it would sound way too squeaky? In other words, am I chasing “good” or “real” sound? For my purposes, I was more interested in the latter, and I had to apply a lot of config and hacks to get there. To briefly summarise, I:

- applied room EQ to the speakers (which needed very aggressive correction)
- matched the reverb the natural acoustics of my living room
- muffled/dispersed the tweeters and coupled the woofer to the drywall behind the NV10S to create soundboard-like dispersion
- put a well-overspecc’d DAC and cable in front of the Kawai’s line-in and spent time optimising gain stages
- Tried different sample rate combinations

It turns out that the built-in speakers are actually very good but not well configured (I tried connecting an external sub but it didn't add much - the built-in woofer is capable for piano sounds). Applying the right corrections helped me produce a convincing organic sound with good mechanical rumble and envelopment. If I imagine a living-room sized grand piano that somehow creates the sound of a concert-grand sized one, I would say the Kawai could fool me in a blind test - it is 90-95% there, way better than I had expected or hoped.

Q: Any more thoughts on the piano models? I continue to think that the Pianoteq SK-EX is exceptional; now that I know how much of an upgrade it is from the built-in sounds, I’ll probably have to get it. The other new development is that I like the U4 more and more - especially with a good speaker system, it sells an upright geometry very well. I still don’t love uprights in general but in the interest of becoming a more versatile player, this might just be my gateway into that world. Playing with the Steinways, I am warming up to them more and more. They have their own voice and it’s been a pleasure getting to know them. In fact, I’m playing them a little more often than the Bosendorfer, which on paper is my favourite. The Bechstein sounds better in the set-up that I have now than it did during my initial evaluation, and I keep flip-flopping between it and my Steingraeber. Jury’s still out on that one. One surprise is that I really enjoy the historical pianos. I can hear their quirks more clearly with the set up I have now, so I can sense more of a connection to the unique instrument that is modelled. I like playing pieces on time-appropriate models as a pallet-cleanser/for inspiration, which is helping bridge the ages a little bit better.

Q: Would you recommend it? Yes, 100%. For my needs, the Novus NV10S is the perfect combination once configured well. I feel I have better access to acoustic musicality with this set up than I would have with with a small, factory-made acoustic grand, which was my initial intention to get. The high-quality keyboard action on the NV10S does the heavy lifting, but a well-configured Pianoteq instance elevates this set-up from an uneasy compromise to a clear winner.

I was thinking about this, and wonder if key release is the right model for this. Since release velocity is just one value at one point in time, it can't represent the exact movement of your key over time, eg. Whether you raise it slowly, or hold it at half release, etc.

If you could map key position to aftertouch (which is a stream of values) that would be better. Pianoteq might already have the controls you need for this, essentially you'd want to instrument damper duration to aftertouch, for each key. The release velocity itself could still be used to scale key release noise and damper noise. That set up should bring you pretty close to the real thing.

I'm not sure how much application this would have in standard repertoire tho. Personally, I'd just be happy to have the damper and key release noises scale with attack velocity, rather than be set to a constant value

Interesting points. I've now adjusted the note-off velocity curve so that it tracks exactly to the note-off velocity value sent by my VPC1. In practice, it seems to make at best a highly subtle difference, not really noticeable during normal playing.

If you could map key position to aftertouch (which is a stream of values) that would be better. Pianoteq might already have the controls you need for this, essentially you'd want to instrument damper duration to aftertouch, for each key. The release velocity itself could still be used to scale key release noise and damper noise. That set up should bring you pretty close to the real thing.

I'm not sure how much application this would have in standard repertoire tho. Personally, I'd just be happy to have the damper and key release noises scale with attack velocity, rather than be set to a constant value

This makes me wonder what the ideal, most realistic [note-off] velocity curve would be.

In a sense it will never be fully realistic as currently implemented, because the program doesn't know the note-off velocity until the note has been fully released. So there will always be a latency. Whereas on an acoustic piano, it's a continuous process -- you can start off lifting your finger slowly and then release quickly for the rest of the key release, and you can hear the note decaying as you lift. In Pianoteq you don't hear the key-release effect until it's finished happening. Or at least that's the way I understand it; I don't have that ability on my NV5s (the only keyboard I have right now) to try it out myself. If Pianoteq one day implements a real-time note release effect one day (probably not possible even with MIDI 2.0 I would think), that would be a dream come true.

Although it's a subtle thing, I think of it this way. What are the things limiting you from achieving the full experience of playing an acoustic piano? The action of your keyboard. The quality of modeling/sampling software. Your speakers. The variables you have at your disposal via different sensors in your keyboard -- key strike velocity (0 to 127 or continuous-valued?), pedal on/off and velocity, resonance effects, etc. And note-off velocity is one of those variables. Whatever progress we're going to see in modeling software in coming years, you can only take advantage of it to the extent that your hardware allows.

The default curve is 127 as the effective key release speed for all note-off velocities (so a flat line at 127). In this case, there indeed isn't any difference in sound for different key-off velocities.

But if you have a perfect diagonal line, you can even have an effective note-off velocity of 0, in which case the key isn't released at all and keeps ringing.

This makes me wonder what the ideal, most realistic velocity curve would be.

All of which is a mouthful to say, for the explicit purpose of non-avant-garde piano playing, I don't feel release velocity is that big of a deal.

I did a test with different key-off velocities for single notes played with identities velocities and I couldn't detect any differences. Perhaps there are microscopic differences in  more complex playing contexts?

I had a similar experience. The NV5s does a great job of reproducing the experience of playing an upright (albeit with a grand piano sound), while the NV10s does a less effective job mimicking the experience of playing a grand piano (for which you'd need a much larger cabinet, with a much more elaborate speaker array).

The action on both is the best I've seen on digital pianos, and the NV5s felt better to me than the Grand Feel III on the CA901, even though this has a longer pivot length than the NV5s. So pivot length in and of itself is not a decisive factor; it's the total design that matters. Both Novus models have just the right sense of inertia, weight and resistance.

I meant which did you like of the audio samples I provided

I thought I settled on 280VC but oscilating between that and Kawai.

Ah, I see. Same here; the Bösendorfer and the Shigeru Kawai work reallly well with this piece.!

dikrek wrote:

I had a choice of one extra instrument when doing the upgrade from stage to standard and went Bösendorfer. The lower register is the most convincing of all models (and I use that a lot).

You can tweak the rest with unison etc. but hard to fix the lower register on other models.

Hey dikrek, thanks for the idea of tweaking the tuning and design parameters. I gave that a spin this evening and managed to bring the Bosendorfer's mids closer to my liking while maintaining the authority of its bass. Pretty happy with the result! Now it feels like I got the best of both worlds.

By the way on the Gnosserie links you shared, I agree that the Kawai and Bosendorfer render the music the best - both have very convincing presence. I'm not all that against samples in principle, but the The Modern D track is the weakest of the bunch to me. The repeated notes sound too similar and their velocity too quantised (I don't think that's solvable with round-robin samples: notes sound different when struck while they are already swinging).

I couldn't say whether The Bosendorfer or the Kawai is better in this case, but I could say the Kawai is better matched to this particular performance: the Kawai's definition helps keep the structure intact during the more freeform bits of play, and because it feels a bit lighter, the accerations sound more plausible.

Thanks. What settings did you change on the Bösendorfer? Try unison at 2.

I had a choice of one extra instrument when doing the upgrade from stage to standard and went Bösendorfer. The lower register is the most convincing of all models (and I use that a lot).

You can tweak the rest with unison etc. but hard to fix the lower register on other models.

Hey dikrek, thanks for the idea of tweaking the tuning and design parameters. I gave that a spin this evening and managed to bring the Bosendorfer's mids closer to my liking while maintaining the authority of its bass. Pretty happy with the result! Now it feels like I got the best of both worlds.

By the way on the Gnosserie links you shared, I agree that the Kawai and Bosendorfer render the music the best - both have very convincing presence. I'm not all that against samples in principle, but the The Modern D track is the weakest of the bunch to me. The repeated notes sound too similar and their velocity too quantised (I don't think that's solvable with round-robin samples: notes sound different when struck while they are already swinging).

I couldn't say whether The Bosendorfer or the Kawai is better in this case, but I could say the Kawai is better matched to this particular performance: the Kawai's definition helps keep the structure intact during the more freeform bits of play, and because it feels a bit lighter, the accerations sound more plausible.