Audio distortion by a speaker may be reduced by detecting onset audio events within an audio signal and modifying the audio to reduce the audio distortion perceived by a listener. The onsets may be detected using a psych-acoustic model by determining critical sub-band powers and corresponding masking thresholds. When a loudness value calculated from the CSBs and masking thresholds exceeds a threshold level, certain frequency bands may be attenuated and other frequency bands may be amplified. The audio modification may be performed on a frame-by-frame basis and each frame may be processed multiple times until the onset is sufficiently masked or attenuated.