Optimum Selection And Repeat Detection Methods Of Audio Example

The development of audio retrieval technology is not a long time, so far research on the audio retrieval technology focused on retrieval algorithms. Example-based audio retrieval is one of the main methods of audio retrieval. In example-based audio retrieval, the selection of audio example has a great impact on detection rate of retrieval system, also a great significance to improve the retrieval system robustness, but very little research on it now. In addition, if there is lots of repeat examples in the system example-library, will lead to a waste of system resources and reduce the efficiency of the retrieval. In order to improve the robustness and the efficiency of the retrieval system, this paper studies optimum selection and repeat detection methods of audio example.In audio retrieval, users tend to submit a longer period of multimedia data as the query input. If make good use of the long data by selecting a segment which having a better robustness to noise as an example, will further improve the retrieval recall and enhance the system robustness. By analysis with different noise robustness of characteristics from audio frames, we can find that the variance of MFCC from audio frames which has better robustness is much larger than the worse ones. This paper rates the robustness of the audio segment by using this feature. Then we are according to the results to select the optimization examples. Experiments show that this method can select optimization examples from an audio segment.In practice, the users always submit a compression multimedia when to select optimization example, so it is necessary to decode it into non-compressed WAV format file. This paper uses DirectShow of Microsoft streaming media solution to decode.If there are a lot of repeat examples in the example-library in the actual operation of retrieval system, not only consumption of storage space, but also a waste of computing resources and will reduce the efficiency of retrieval system. So it is necessary to detect of repeat examples. Then make the appropriate treatment if finding repeat examples. Using of segmentation-based audio retrieval algorithm in detecting of repeat examples which can speed up the retrieval speed. By setting of three matrices to record search results in segmentation-based audio retrieval algorithm, we can obtain the similar information between the examples after retrieving. The experimental results show that the method can quickly identify repeat example and can remove repeat examples according to threshold set by user.