The potential of speech as a biomarker for health has been realized for diseases affecting respiratory organs, such as the common Cold, or Obstructive Sleep Apnea, for mood disorders such as Depression, and Bipolar Disease, and neurodegenerative diseases such as Parkinson’s, Alzheimer’s, and Huntington’s disease. The recent progress achieved with machine learning methods trained with lab data recorded in very controlled conditions has been impressive, but it also raises questions of robustness and privacy. The first part of this talk addresses the robustness question, discussing how these methods perform with in-the-wild data, concretely, with data from vlogs collected from YouTube of subjects potentially affected by one of the aforementioned diseases. The second part addresses the privacy concerns posed by the possibility of mining speech signals for health related cues, in a world where speech data and the info that may be extracted from it may be legally regarded as Personable Identifiable Information.