Audio Segment-based Audio Retrieval Clustering Algorithm

With the further development of internet and search engine technology, searching on the data of the webpage's text can not be satisfied with people. So multimedia retrieval technology, especially with audio retrieval, has become to be a hot point of the research. And it's also an important developing direction of internet technology in the future.Content-based audio retrieval is still a difficult problem of research, two major methods with text-based audio retrieval and phonetic-based audio retrieval are widely used. The former is based large vocabulary continuous speech recognition and traditional text search technology; the latter comes from the feature of audio and signal processing. With phonetic-based audio retrieval method, we use traditional methods of feature-based retrieval and lattice-based retrieval as the baseline and talk about their advantage and shortcomings. Then we introduce a new algorithm with audio segment based clustering algorithm. This algorithm is accurate and highly effective especially with search engine. So it has some theoretical value and prospects. With text-based audio retrieval method, we focus on the combination between text-based retrieval and phonetic-based retrieval. And we introduce the combination algorithm and function based on audio segment-based clustering algorithm. In this way, it expands the applicability of the new algorithm.Now we don't only focus on the ranking of the search results but also the clustering. At the same time of analysing the primary algorithm, we introduce a novel search results clustering method named search-based text clustering algorithm, the algorithm can be say as a creative use of the search engine.Reference case finding system is an audio retrieval system based on these key algorithms. This system focus on telephone data and contains an integrated audio search engine based on phonetic and text combination and a tool which can be used to do the search results clustering. Via test, this system has a good performance and efficiency, and it has achieved a good rate of accuracy and speed of search as a good search engine. So the practice proofs that these algorithm is feasible and availability.