Voice AI technologies are increasingly promoted as biomarkers for detecting and monitoring health conditions, yet standards for evaluating their reliability remain underdeveloped. This chapter argues that voice AI quality must be judged not only by performance metrics but by the sufficiency of evidence, from model validation to clinical and ethical accountability. We highlight how common practices such as data leakage and limited cohort diversity cold distort claims of clinical readiness. Drawing on emerging regulatory and ethical frameworks, we push for a multidimensional view of quality, grounded in technical robustness, clinical utility, and societal alignment. Voice, as both a biometric and behavioral signal, exposes gaps in current AI evaluation norms and demands domain-specific standards. The chapter positions rigorous, transparent evaluation as a prerequisite for trustworthy and equitable deployment of voice-based health technologies.