How do listeners sustain meaning when speech is masked by noise? I address this using the N400, a component of event-related potentials (ERPs), to test predictive processing during spoken sentence comprehension. Analyses of N400 amplitude and latency show how contextual predictions mitigate the impact of degradation and how these dynamics evolve during sustained listening. I also use the N400 as a biomarker for evaluation: across clear, noisy, and enhanced speech, shifts toward a clean-speech profile index more efficient semantic integration and provide a theory-grounded benchmark for comparing enhancement methods. In an independent line of work, alpha-band (8–13 Hz) activity indexes listening effort. Lightweight, subject-efficient classifiers convert alpha features into a continuous effort score that complements intelligibility and preference ratings, identifying cases where signals are understandable yet cognitively taxing. Together, N400 and alpha provide complementary axes—semantic efficiency and cognitive load—supporting rigorous, EEG-driven evaluation of speech enhancement under adverse listening conditions.