Abstract: This paper presents a segmentation-based real-time frequency retrieval method and discusses its control strategy. In the method, the retrieval target is divided into a series of segments and a retrieval window is used to control the search of segments. This method can obtain a very high retrieval speed that is independent on target length and can be controlled by the size of retrieval window. In multi-target retrieval, audio classification is used to reduce the computation of similarity. Experimental results show that recall rate and accuracy rate of retrieval method can achieve 100% and 99.7% respectively and the method can maintain high performance even if large part of the target is absent in the input stream. Classification can effectively improve the speed of search and the average accuracy of classification is 95.7%. Retrieval response can be triggered with little time lag. It is suitable for real-time application to retrieve audio information of any length from unknown input data.