Synth Mapper

Computational Creativity Final Project

Audio Results

Audio Reconstruction

These results show the sound matching results with respect to a reference target sound. The target for a number of different sources is shown. The genetic algorithm (GA) result is the best individual from the last population generated during optimization. The variational auto encoder (VAE) result is the audio generated from reconstructing the parameters used to generated the GA result.

Timbre Interpolation

These results show audio generated by monotonically varying one dimension of the latent space of a trained VAE, which was regularized on a timbre feature. The starting point is the latent vector representing the best individual for each sound generated by the GA (same as above). A single dimension from the latent vector was modified by adding linearly increasing values from \(-1.0\) to \(1.0\) and synthesizing the result. All other latent dimensions were held constant.

S. refers to spectral and T. refers to temporal.