Moderate but not high noise during training is associated with stronger generalization to quiet

Speech learning often occurs under adverse acoustic conditions, yet learned representations are later accessed in quiet. This study asked whether word learning acquired under background noise generalizes to quiet listening and whether generalization depends on training noise level. Sixty young adults trained within a 5-day pseudoword-picture association task under Quiet, +10 dB signal-to-noise ratio (SNR), or 0 dB SNR speech-shaped noise. Learning was assessed after each session using an identification task conducted in quiet. Accuracy improved across sessions in all groups. When tested in quiet, performance following +10 dB SNR training matched Quiet, whereas performance following 0 dB SNR training was lower. These findings suggest that generalization to quiet depends on training SNR, with robust transfer following moderate noise.