Abstract: A improved speech voiced/unvoiced classification algorithm based on the band-partitioning spectral entropy and Teager energy of wavelet transform is proposed in this paper.First, the band-partitioning spectral entropy is computed.Second, the Teager energy of every band of speech decomposed by wavelet transform and the ratio of the energy in the wavelet low-bands to total energy of each speech segment. Finally, the voiced and unvoiced decision is made according to these two parameters.Experimental results show that the proposed approach outperforms the amplitude energy-based algorithm.