Abstract: Effectively extracting feature information of speech signal is the key of speech recognition. By empirical mode decomposition method, a series of intrinsic mode function can be got.Extraction process of intrinsic mode function is to reduce its redundancy. Experimental results show that the empirical mode decomposition method can decrease the recognition error rate. EMD is an effective method. It is feasible to extract the feature from speech signals with EMD.

Key words: Speech signal, EMD, Feature information, Speech recognition, Intrinsic mode function