海量音频指纹数据的存储与检索研究-计算机科学与技术专业论文.docxVIP

ABSTRACT With the coming age of big data, the world is producing data at an exponentially increasing speed, especially multimedia data such as images, audios and videos. How to effectively manage and make use of these data to providing more convenient is one of the fundamental problems people need to solve in the information age. As the development of techniques in pattern recognition, machine learning and cloud computing, content-based multimedia search comes in. Compared with traditional keyword-based search, content-based search is independent of tags and keywords, and with more accurate search results and more convenient search methods. As the important component of multimedia data, the data amount of audios also increases fast. The key problem of people faces is no longer lacking of data, but how to find data they want in massive data. And how to retrieve audios from large-scale databases effectively and efficiently is a big challenge for both academia and industry. Audio fingerprinting technologies is one of content-based audio search methods. By extracting digital features called audio fingerprints from the unknown audio segment, and then search and calculate similarities in a prepared audio fingerprint database, we can get detail information of that audio. This method avoids the problems such as lack of tags or have wrong tags exists in traditional keyword-based search. And at the same time, this method could help users find what they want even when they don’t know the keywords. The algorithms of audio fingerprints extracting and matching have achieved significant results in some laboratories, and have been applied to some products with relative small datasets. However, large-scale datasets always introduce performance bottleneck, and problems about concurrency and extensibility. This paper designs, implement and optimize the storage and retrieval of massive audio fingerprints based on the deep research in algorithms of audio fingerprints extracting and m