In the context of machine listening, the spectral appears in, with, and as the voice. This contribution asks how the voice can announce the body—specifically, how the body gets (re)constituted as a referent to knowledge through the spectral deconstruction of the voice and its subsequent manipulation as spectralized data. The framework for this understanding stems from my ongoing scholarly and artistic work around the so-called dialect recognition software, a proprietary solution used by the German migration authorities since 2017 in the case of undocumented asylum seekers. Through specific artworks and a theoretical discussion of their processes, the contribution seeks to challenge the assumption of an immutable “a priori” connection between body and voice, instead placing it in the realm of listening, machinic or human.