<p>With the advent of datafication and AI-mediated analytics, computer scientists and health researchers have begun to target voice data as potential biomarkers. This paper follows the technoscientific and speculative strategies for deploying voice data as biomarkers in health research. Contributing to infrastructure studies, biomedical science studies, and social studies of AI, we describe and analyze emergent knowledge infrastructures involved in translating and enacting the human voice into a set of biomarker variables. This includes start-ups pursuing standardization of voice samples as referential ‘ground truth’ datasets and developing AI techniques as products for clinical trials. Conceptually and strategically, developers of vocal biomarkers have coined the notion of the ‘voiceome’ to draw on omics infrastructures. They use those existing omics platforms as available templates for how to proceed to study voice features as markers of health and disease. Adding machine listening to digital phenotyping and precision medicine, vocal biomarker research may transform psychiatric practices in specific ways under certain erasures and anticipated value, building on infrastructures that are not (entirely) new, but rather complemented and afforded by automation and predictive screening.</p>