During my two final years of undergrad, I worked at Ithaca College's music innovation lab: JimiLab.
I worked in a number of areas related to music technology and machine learning, including recommender
systems, generative models, and DSP.
You can view the details of these projects here:
Content-based Music Similarity with Triplet Networks
Synthesis with Generative Models
VST Plugin - Delay-line based synthesizer
My first summer at JimiLab, in collaboration with Arshiya Gupta, we embarked on a project to estimate
synthesizer settings from audio. The idea being that one could feed in a sample from their favorite
synth-pop song, and
automatically setup their synthesizer so that they could play along.
We ran many experiments with Convolutional Neural Networks, among other ML architectures, to predict
parameters
used to generate samples from
many different synthesizer types, spanning FM, Additive, Subtractive, and Physical Models. Along the way we
implemented an FM and Karplus-Strong synthesizer to
generate training data.
Although we did not have great results with our estimator, this project gave me a really solid foundation in
machine learning and DSP, which
contributed to greater success with the projects listed above the following year.