Hacker News Reader

Comment by olau | Hacker News Reader

A little nitpick on the sound quality of piano VSTs: I don't know about drum VSTs but I don't think your piano VST is actually a good example, because piano synthesis is actually pretty darn hard.

I haven't followed this closely the past few years, but it used to be pretty obvious you were listening to a VST compared to a real instrument in actual playing. When you compare recordings, sample based ones sound awesome as long as you just try a single note, but they were obviously missing something in the sympathetic resonances (sounding worse than my upright piano which is perhaps a 20th percentile piano), and the more physics based ones had not so great tones.

I've actually been hoping that someone would try to apply deep learning to piano VSTs, even emailed a few researchers, but no reply. If something new has come up I'm not aware of, I apologize.

View on HN

Have you checked out Pianoteq? It is physical modeling and sounds incredible. It even runs on a RaspberryPi of you want.

When it comes to real instruments, Roland ships physical modeling in some of their instruments. They are a mainstream company.

So I think physical modeling is there but harder to get right than recording samples of piano keys being pressed. The results can be been convincing and it's nice that it requires little memory compared to sample based techniques.

I recall seeing a audio technician test some of the modern VSTs and it seemed extremely impressive to me as an amateur.

It's to the point where you virtually put microphones around the virtual instrument and it behaves (according to the person who does it for a living) creates an intuitively correct sound.

Edit: and to be clear this was a demo from before the LLM times so it's something built the traditional way.

I get where you're coming from, but IMO, for the 99.99% of modern music where either the piano is sitting in a busy mix with other instruments and/or the piano is getting routed through a ton of compression, saturation, delay, reverb, etc to make it sound like a modern track, you really won't be able to tell whether it's a VSTi or live recording. True for both pop music and soundtracks. The only case anyone can tell a difference is going to be solo piano pieces that are intentionally made to sound as raw as possible, which is fairly niche.

You can bring this distinction back to recomps, too -- one area you can't use a vibed recomp over an emulator is things like TAS's[0], since the TAS community has strict rules about staying compatible with the original hardware, including doing things like sending the TAS inputs into an actual console to verify it works properly. A recomp is likely going to have minor (or major) compatibility issues that would prevent this from working. But it won't matter for 99.99% of players.

[0] https://tasvideos.org/

Reply on news.ycombinator.com