A wind-noise suppressor based on wind-onset detection and spectral gain modification is proposed. Wind onsets are detected based on an estimated SNR to find frequencies of sharply increasing components in each frame of an input noisy speech. The detection result is modified by speech presence likelihood based on spectral smoothness to eliminate speech onsets. To suppress wind noise with little speech distortion, spectral gains are made smaller at the detected wind-onsets. Subjective evaluation results show that the 5-grade MOS for the proposed wind-noise suppressor reaches 3.4 and is 0.56 higher than that by a conventional noise suppressor with a statistically significant difference.