Recently, active networks have been highlighted as a key enabling technology to obtain immense flexibility in terms of network deployment, configurability, and packet processing. Exploiting this flexibility, we present an active network application for real-time speech transmission where plugin modules are downloaded onto certain network nodes to perform applicationspecific packet processing. In particular, we propose to perform loss concealment algorithms for voice data streams at active network nodes to regenerate lost packets. The regenerated speech data streams are robust enough to tolerate further packet losses along the data path so that the concealment algorithms at another downstream node or at the receiver can still take effect. We call our approach active concealment for speech transmission to distinguish it from concealment performed at the receiver. Our approach is bandwidthefficient and retains the applications' end-to-end semantics.