Centrifugal pumps are widely used in various fields. The cavitation not only causes the destruction of over-current components so as to influence the reliability of operation, but also affects the pump running stability because of vibration noise. In many cases, cavitation is unavoidable. Therefore, it is necessary to improve the accuracy of cavitation monitoring and reduce the unnecessary loss caused by cavitation. The objective was to study the variation law of noise characteristics under different cavitation conditions of centrifugal pump, and the influence of cavitation development on hydrodynamic noise. The previous scholars used the test methods to get the noise signals, but the signals were not the same as different product characteristics of pumps. Therefore, it is a cost-effective method to predict the cavitation noise by numerical simulation. And the experimental method was used to verify the accuracy of numerical calculation. Firstly, an ultra-low specific speed centrifugal pump was used as the research object to build a closed test bench. Based on the pump product test system and data acquisition system, a test system of centrifugal pump cavitation noise and performance was established to realize pump performance parameters and internal field noise signals synchronization acquisition. The cavitation performance curves of the model pump were predicted by Kunz cavitation model and Zwart cavitation model respectively. And the experimental values were compared with to select the appropriate cavitation model. According to the vapor volume fraction distribution among blades and cavitation performance curve, the whole cavitation process was divided into noncavitation stage, nascent cavitation stage, feature cavitation stage and serious cavitation stage. The effects of cavitation on internal flow field and pressure at different times during pump operation were analyzed. The acoustic boundary element method was used to transform the flow field information into sound field information. The relative error of the noise prediction value and the experimental value of each cavitation stage was compared to verify the feasibility of the forecasting method. Because the analog signals show high degree of coincidence with the actual signals. Finally, the influence of cavitation on the internal sound field was studied based on the flow sound coupling method. The study showed that for the cavitation on ultra-low specific speed centrifugal pump, Zwart model had better applicability compared with Kunz model. The relative errors between the predicted value and the experimental value of the Zwart model were all within 5% in the non-cavitation stage, the feature cavitation stage and the serious cavitation stage. The cavitation deteriorated the flow regime in the flow channels, especially from the feature cavitation stage. The number of internal vortices was increased and pressure fluctuation was increased because of the vapor, which made the increase of instability. The variation of the inner field noise signal with cavitation was complicated. In the low to middle frequency range, the sound pressure level of the discrete component of blade passing frequency and its harmonic frequencies were gradually decreased with the development of cavitation, due to the inhibitory effect of cavitation on dynamic and static interference;while the axial frequency component showed an increasing trend. The sound pressure level in high frequency range was decreased firstly and then increased sharply with the decrease of the cavitation coefficient. The high frequency eigenvalue component was gradually submerged in the wide frequency band. The increase of the sound pressure level in the high frequency band caused the total sound pressure level risen. Compared with the experimental results, the relative error was less than 5% in the non-cavitation stage and nascent cavitation stage. In feature cavitation stage and serious cavitation stage, the degree of coincidence was slightly worse, and the relative error was less than 10%.