Ultra-high pressure water jet nozzle speed is the important factor influencing the water jet cutting quality and efficiency. Different sound characteristic value will be generated when cutting different materials or geometric shapes of the object and the best match of ultrahigh pressure nozzle movement speed will be reflected by it indirectly. The water jet cutting voice signal acquisition system is designed in this paper. Data process of the cutting sound signal is based on second generation wavelet transform and the Teager energy operator, and the quality of cutting workpieces is got. Using the method of least squares vector machine of small sample, the nonlinear model is established between the movement speed and the sound feature, and the adaptive speed optimization control of water cutting is realized. Finally, the effectiveness of the proposed method is verified by field test.