Virtual Instrument Jazz Performance Crosses The Uncanny Valley

In this video, Roger Gentile performs his virtual instrument arrangement of the jazz standard, When I Fall In Love, by Victor Young.

While percussive sounds have long been sampled or synthesized with reasonably realistic results, until recently, attempts at creating realistic solo brass performances have often sounded cheesy or ‘uncanny valley’ off.

In this performance, though, Gentile plays Audio Modeling’s SWAM Flugelhorn – a physical modeling instrument, running on an iPad- creating a performance that’s very realistic and, more importantly, musically expressive.

Here’s what he shared about the technical details:

In this recording, the following instruments were used:

Swam Flugelhorn, Yamaha KX88 with BC3, Midisolutions Breath controller, CME Widimaster, Hammond B3 and GSi Burn.

SWAM (Synchronous Waves Acoustic Modeling) is a synthesis approach that combines concepts of Physical Modeling and Behavioral Modeling. Swam Flugelhorn uses Audio Modeling’s SWAM-B engine, which synthesizes instruments using pure Physical Modeling, so there are no samples used.

Check out the performance and share your thoughts on it in the comments!

14 thoughts on “Virtual Instrument Jazz Performance Crosses The Uncanny Valley

  1. Very nice model and performance. What interests me the most about physical models is taking a model like this and pushing it beyond just emulating the natural response of the instrument. SWAM are very focussed on emulating the real, I would like more access to the parameters of the model so I can use it for more exploratory synthesis.

    1. One of the gotchas of physical modelling is that tweaking parameters doesn’t give you anywhere as much flexibility as you might expect, ie: a physical model of a trumpet is going to sound like a trumpet, or may be you’ll degrade the sound, or it will be unpredictable, or maybe it won’t play at all. There aren’t many sweet spots, nor will it ever sound like a saxophone, let alone a violin or a guitar which would require different models.

      1. Respiro both allows you to explore physical modeling into more exotic territory, but it also does a good job of keeping the instrument relatively stable. It’s one of the most musical synths I own.

    2. Same here. There’s still a branch of PM which goes in that direction. My current favourite is IMOXplus Respiro. I’d say it’s a spiritual successor to Yamaha’s VL line. Unlike the VL, it doesn’t actually give you access to all the parameters in the model. It does give you access to several macros. And several of its models sound like “impossible acoustic instruments” in that they sound acoustic yet unlike any instrument you know.

      The SWAM Engine is definitely not going in that direction. For instance, the software “decides” on articulations instead of giving the player full control over articulations.
      Talking with Audio Modeling, I get that we’re not their core market. One way to put it is that they optimize for realism instead of performability. The way I’d describe this is that it’s more for keyboard-based scorers than for wind players.
      Still, they’re open to the idea of releasing other apps which would go in other directions. It might take a while, given their current focus on expanding their line within their core market (woodwinds for iPadOS, for instance).

  2. It’s tragic that we can’t get hold of the BC3s etc anymore, instead you have to buy a fancy thing with usb etc.. Using a breath controller is a lot of fun, and possibly more fun playing than listening back.

    1. Yamaha’s unlikely to get back into breath controllers… or Physical Modelling. While it’s still possible to buy BC3-like devices from third parties, they typically work with adapters, especially USB-A ones. So, yes, you end up having to buy a fancy thing with USB, like the TEControl “USB MIDI Breath and Bite Controller 2”.

      One promising development is the Photon mouthpiece. It’s partly led by Rudy Verpaele, the dev behind the Respiro Physical Modelling softsynth and some editors for the VL series PM hardware.

      1. Thanks, I am watching the Photon closely! It’s just a shame because I have several old synths with BC inputs and BC3 looks like a fairly simple technology. Might try a DIY at some point.

  3. For me all their instruments have a very short wow effect, but after that it sounds very fake, like a synth (which it is). It sounds good and dynamic and lifelike until you turn on a track with the real instrument. Brass are somewhat decent, tho, the strings are the absolutely worst. But I would be happy to see these plugins reach “if you can’t tell – does it matter” level of authenticity in my lifetime.

  4. The instrument is clearly very expressive and sounds very authentic to my ears, but it really feels like the breath-controller is the key to making this work here, would be hard to do this with a modulation wheel for example.

    1. The only way for a performance to sound ‘real’ is for the performance to be real – regardless of whether it’s on a traditional or virtual instrument.

      So using mod wheels to simulate breath control (or vibrato) always sound fake, because its a different physical gesture. To sound real, you need that real breath performance.

      It’s the same way with vibrato & string synthesis – using a mod wheel immediately makes it sound phony.

      It’s funny how these same things sound fine with ‘synth sounds’, but as soon as you try to do something realistic, it sounds cheesy.

  5. This is a straight up mind-f*ck!

    I immediately forgot that this was ‘virtual’ and found myself just enjoying a great rendition of a standard.

    I doubt that many jazzers would be able to identify this as a ‘synthesized performance’.


  6. That’s a lovely performance and a solid pitch for SWAM. I’d like to think that a few brass players will use this for practice in places where woodshedding with the real thing would get them evicted.

    My first question is: How many people will take it up in a bid to sound “real” vs. those with experimental goals? I’d like to see the total sales numbers, because its so esoteric.

    Second: How many players will have (or develop) the required “BC2” chops to hit either mark? Its like using MPE. Convincing casual listeners and convincing musicians aren’t always 100% the same thing.

  7. Not bad, the spectral content of the attack transients are a little too consistent, which makes it feel like a synth.

  8. I think for non-players of those instruments, the SWAM instruments have a striking realism. As a trombone player, I listened to their trombone instrument and on some level it was “impressive” but it was not only fake sounding, it was a mildly annoying. “A human trombone player would NEVER do that.” I think the uncanny valley description is apt.

    I think the SWAM approach is good, and worthwhile for some. But as David Bessell mentioned above, it can be more interesting to explore new territory.

    In addition to Imoxplus’s excellent Respiro instrument, Reaktor also has something called Silverwood which is pretty nice.

Leave a Reply