This paper describes an on-line, end-to-end and non- intrusive approach to estimate the vocal transmission quality over wireless ad-hoc networks. The proposed assessment algorithm, denoted PEVOM (perceptual evaluation of voice over MANETs), is based on the fact that the observed quality of packet based voice conversations is time varying and that mobile users may tolerate degradation of perceptual quality and even the interruption of service. The evaluation of transmission quality adopted by PEVOM is relying on e-model paradigm. PEVOM includes an original extension to e-model to adequately rate the distortion effect due to the intermittent nature of established connection incurred by mobile users over a MANET. Specifically, PEVOM checks passively the connectivity between communicating parties by monitoring periodically the ratio of active to total time. When mobile users are connected, PEVOM calculates the perceptual quality using the conventional e-model. However, when mobile nodes are disconnected, PEVOM calculates the perceptual quality according to an expected recency effect. At the end of each connected/disconnected period, PEVOM estimates the perceptual quality of each interval using a weighted average over time. At the end of a vocal session, PEVOM rates the overall conversational quality using a weighted average over time of the produced rating factor of each connected/disconnected interval. The rating behavior of PEVOM is verified through a study of the vocal transmission quality over an intermittent multi-hop wireless connection.