Machine Learning Viola
Instrument made by feeding piano samples into a machine learning model that tries to replicate the timbre that a viola produces
BySam Knight
The story
I’ve recently gotten interested in machine learning for music applications due to a class I’m taking at school. I’ve specifically been messing around with timbre transfer using the Magenta DDSP library. This sample instrument was made by feeding piano samples into a machine learning model, that was trained on 10 minutes of viola playing. It then takes those piano samples and tries to make them sound as close to a viola as it can. I thought it created very cool artifacts that significantly added to the sound, so I put it in Kontakt right away. Enjoy! And thank you to Noah Leong for the viola model
Reviews
Incredibly unique sound, exactly want I want
I feel like people aren't understanding the uses for this pack. It seems to me like the point isn't to have a usable string library, but instead something with many unique applications outside of replicating strings. I personally love the artifacts it creates, especially in the upper end. Especially when playing around with the ADSR and filter, it can just sound like an interesting synth. Really excited to see more like this.
SNES strings
What a cool concept! It has this Super Nintendo crunchy vibe about it, and I particularly like the instability in the higher register. Would love to hear more of what machine learning can come up with!
Interesting Concept
Really love non-traditional sample instruments and this may be the most interesting idea I've come across. I love how the tail of each sample starts to glitch and gives it a really interesting timbre. I found the most usable results by bringing the attack back a bit and rolling the cutoff down for a nice pad. It lessens a bit of the high end harshness that is inherent in the sample.