Presenting a Poster on DDSP Source Separation at SANE 2026

I’m excited to be attending SANE 2026 (Speech and Audio in the Northeast) at MIT in Cambridge, Massachusetts on October 30, where I’ll be presenting our poster, Discrete Parameter Estimation for DDSP Source Separation, with Andrea Cogliati. The work looks at how to fit the integer parameters of a Karplus–Strong string model, its loop delay and pluck position, where gradient descent breaks down, and compares classical search and pitch estimation with relaxed and learned categorical estimators (REINFORCE, Gumbel-Softmax) on mixtures of up to seven strings. SANE is a one-day workshop that brings together speech and audio researchers from across the Northeast.