In this paper, we present first evidence for a potential application of novel speech technological methods as a valuable tool for basic phonetics research. We describe a research program aiming at identifying the complex phonetic realizations underlying various dimensions of phonetic variation. This will be addressed with the help of recent approaches in unsupervised voice conversion and waveform generation. Concretely, we present a model for disentangling speakers' voice qualities and their linguistic-phonetic content, which can then be used to perform voice conversion across different dimensions of phonetic variation. The resulting signals are then "audible versions" of the phonetic dimensions of interest, and lend themselves to straightforward phonetic interpretation.