Looking for a substantial increase in the reliability and validity of voice analysis measurements in depression, we have developed a new experimental paradigm. The paradigm is based on two vocal tasks with different levels of cognitive processing and a baseline verbal task. It has been applied to 40 subjects, 20 patients with affective disorders and 20 normal individuals. We have found that an important set of vocal parameters, at different intervals, discriminates patients from normal subjects. Here, we discuss four of them – those that have been especially consistent; the patients had longer reaction times, a higher proportion of relative spectral energy (at a very low frequency band), narrower fundamental frequency range and higher proportion of omissions. Under the proposed paradigm, the acoustic analysis could be a promising instrument for the development of quantitative indicators of depression.