Abstract: A novel algorithm of finding the speech formant frequency by using the character of Hilbert-Huang transform as an adaptive band-filter is proposed.The algorithm utilizes the character that the intri-nsic mode functions,which are zero mean narrow band frequency-amplitude modulation signals,are consistent with the frequency-amplitude modulation model of speech perfectly.Besides,according to the adaptability of empirical mode decomposition,the character that the signal itself decides the frequency and band of intrinsic mode functions is effectively applied.Thus the formant components can be obtained with-out having to estimate the frequency and band of the band-filter in advance,and the influence of the interference of false peaks,merged formant and high-pitch voice can be avoided.This algorithm not only can accurately extract the ormant,but also can track the changes of the formant.