This paper addresses the impact of packet-based transmission on the performance of speech recognizers and on the quality of synthesized speech. For both cases, degradations which are typical for packet-based networks have been generated in a controlled way, using a parametric simulation model. Recognition performance measures and subjective quality judgments are compared to the quality degradation which is expected in human communication over such channels. The results show that the impact on ASR performance differs from the one on speech quality. The latter may be predicted with a quality prediction model.