Abstract: Presents a 2.4kbit/s characteristic waveform interpolation (CWI) speech coding algorithm based on Haar wavelet lifting. CWI amplitude spectrum obtained from discrete time Fourier series (DTFS) is firstly converted to discrete cosine transform (DCT) coefficients, which are then decomposed and reconstructed by Haar wavelet lifting. Voiced degree is decided by the mean value of phase spectrum distances and pitch cycle prediction. It is used to select suitable phase model and the selective quantization of DCT coefficients. Subjective A-B listening test showed that the quality of the proposed speech coder is better than that of the traditional 3.8kbit/s CWI coder. The problem of long-delay CW decomposition and reconstruction with traditional wavelet transform is also solved by the Haar wavelet lifting scheme.