Study On Content Based Audio Information Retrieval

In recent years, with rapid growth of the multimedia data, Content Based Multimedia Information Retrieval becomes more urgent. Content Based Audio Information Retrieval technology is important in a considerable number of areas, such as remote teaching, health care, digital libraries, environmental monitoring, search and entertainment news editing and production. This paper carried out exploratory research around the two branches of the Content Based Audio Information Retrieval:Content Based Speech Retrieval and Content Based Music Retrieval. Main tasks can be grouped into the following three parts.STD retrieval based on syllable confusion network. First, convert broadcast speech documents to text in confuse the network format by speech recognition technology. Then use text retrieval technologies to find the speech documents related to the input keyword query and return the in order. We analyze the effect of different pruning strategies to system performance through the experiment.MIDI music retrieval query by humming. First analysis the format of MIDI, extract the melody. Then extract the melody fragments of humming query by pitch extraction algorithm and calculate the similarity between the query and database. At last return results order by the similarity. We analyze the effect of different match algorithms to system performance through the experiment.Audio information retrieval based on audio fingerprinting. First use graphics related algorithms to extracted feature points from spectrum. Then use hash structure to find matching feature point and return the similar audio. We analyze the effect of different index structures and feature extraction algorithm to system performance through the experiment.Finally, summarize the whole paper, and look into the distance to the hot spot and the systematic developing trend on CBIR.