| With the development of Internet, the content and form of information processing has been more and more various. Multimedia processing plays an important role in information processing area. Audio as one of the main form of multimedia occupies an important position in people’s daily lives. Extracting interested audio clips from massive audio files has a wide range of application in multimedia retrieval, intelligent monitoring and other areas. But mass audio data needs more efficient storage system, and the processing of classification and retrieval on audio data is a computationally intensive task, which stand-alone platform can not undertake. As the distributed computing and storage technology has been more mature, the distributed system such as Hadoop has been used more and more widely. Cluster composed by the inexpensive PCs has been able to meet our high throughput, high performance requirement.Based on the research of audio retrieval and distributed computing technology, we design and implement a content-based audio retrieval system. Our work includes audio library creation and content-based audio retrieval. The former one transforms audio files with different sizes and different format into unified WAVE audio format, and put them into the distributed audio library. The audio retrieval function based on the characteristics of supervised learning, using two distributed tasks—the training task and the retrieval task to achieve content-based audio retrieval job. At last, we deploy the system into a cluster composed by five PCs, and verify the effectiveness and performance in two practical application—analysis of monitor audio data and pornographic information detection. |
