In this paper, we propose a low bit rate waveform interpolation speech coder where the novelty lies with an effective decomposition method of pitch cycle waveform(PCW). PCWs exhibit very different perceptual characteristics in different frequency bands. For frequency components below 1kHz, they are quantized using Variable Dimensional Vector Quantization(VDVQ) scheme. Hereby retaining the fine harmonic structure of the speech signal. For the upper frequency band ranging from 1 to 4kHz, the formant structure is perceptually more dominant. It is therefor desirable to capture the formant peaks in order to maintain high speech quality. In this case, we employ a sparse frequency representation(SFR) method. A 1.7kbps speech coder has been developed by this technique and the quality of the output speech is perceptually good.