Parametric spatial audio rendering promises fast and perceptually convincing audio cues that remain playback-system agnostic. We propose a parametric encoder for spatial room impulse responses that is tested with nine simulated rooms spanning a large range of sizes and reverberation times. A key component of the pipeline is a perceptually inspired model for determining a minimal set of salient early reflections to reduce computational complexity. The results of a listening study with 27 subjects suggest that rendering six early reflections is indiscernible from a fully-rendered reference for the tested speech content and frequency-independent room simulations based on the image source method. However, the proposed model requires further improvements with respect to detecting and selecting the most-salient early reflections.