This paper describes our efforts at characterising the caregivers voice in retirement homes. Towards this aim, we recorded 20 professional caregivers in two retirement homes. These caregivers were recorded using a headset microphone connected to a smartphone, allowing a total liberty of movement while keeping a very high sound quality. The caregivers were recorded while performing three different tasks: text reading, informal interview and professional roleplay with a fictive patient. All the recordings were then processed by an automatic speech recognition system that gives words and phones sequences and their timestamps. In our analysis, we focus particularly on the differences between spontaneous speech conditions and look for contrasts in terms of duration, speech rate, fundamental frequency and intensity. Our final aim is to capture the specificity of the professional caregivers voice in order to build automatic training tools.