Speech intelligibility in adverse situations, such as reverberation and noise, is conserved until the degradations reach certain thresholds. Psychoacoustic studies have described the properties of speech that lead to the conservation of its intelligibility under those circumstances. The neural mechanisms that underlie the robustness of intelligibility in these situations, however, are not yet well understood. Here, the cortical representations of speech in reverberation and speech plus noise in reverberation are studied by measuring the cortical responses of human subjects using magnetoencephalography (MEG) while they listened to continuous speech narratives. It was hypothesized that the neural processing of speech in reverberation and speech plus noise in reverberation would follow a lack of cortical synchronization as function of the degradations. Encoding models show, however, that the neural encoding of speech in reverberation follow a different mechanism than that of speech in noise. On the other hand, in the absence of noise, it is possible to reconstruct with high accuracy the envelope of reverberant speech, thus demonstrating that the reverberant speech is well encoded by the brain.