Hacker News new | past | comments | ask | show | jobs | submit login

Everyone is writing about their startups. My startup makes self-driving delivery robots (http://robby.io).

But that aside, one of the things I'm working on is trying to use RNNs to create a better digital piano. Even the best digital pianos out there are far inferior to a even a YouTube recording of a concert grand piano. One of the biggest problems I notice is the complete decoupling of resonances between strings; most digital pianos treat notes independently and just sum up the audio signals. In reality it's a giant, heavily interconnected physical system with tons of resonances and nonlinearities, and I want to see if some signal processing combined with backpropagation can be used to abuse a neural network to simulate the energy transfer in a physical system of that complexity.

I haven't been terribly successful yet, but it would be amazing if there existed an open source digital piano that performed spectacularly and could be plugged into an el cheapo weighted keyboard for decent piano sound.




Do you know pianoteq?

https://www.pianoteq.com/


Thanks! This is interesting.


Pianoteq is without a doubt the best sounding piano synth on the market today. Uses all sorts of interesting physical modeling algorithms (including string-to-string resonances, with a more-than-first order model...), unfortunately, unpublished. ;)

If you're interested in chatting about sound generating software and algorithms, feel free to shoot me a line. "Also, do I remember you from ec-discuss?"


Yep I used to live at EC :)


Are you sure that you are comparing with the best digital pianos? I mean they have been working on more life-like digital pianos for decades, going back at least to the Kurzweil stuff https://en.wikipedia.org/wiki/Kurzweil_Music_Systems




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: