Research And Implement On The Content-Based Audio Retrival

Audio information is a kind of very important multimedia information and it's more and more significant to quickly and effectively search our desired information from audio information. However, audio information retrieval techniques didn't attract enough researchers; especially real-time content-based audio information retrieval has been rarely mentioned.This thesis discusses the so-called content-based audio retrieval in real-time based on multiple patterns. Its researching object is television advertisement and the aim can detect given television advertisement from the playing television programs, and the monitoring process is real-time. Through the result the advertisement beginning time and ending time can be located, and the advertisement times can be known.This thesis, according to the characteristic of real-time audio data of television signal, gives an overall design scheme. The scheme takes two steps. The steps are frame sequences matching technique and locating the templates of advertisement. That is to say locating the advertisement starting part from the current real-time audio information firstly and the judge whether the audio data is the detected television advertisement according as locating result. According to the characteristic of real-time audio data of television signal, chooses the effect audio character for detecting the advertisement from the audio retrieval, utilizes the audio character and detecting the templates of advertisement head and cuts the advertisement length as the detected templates of advertisement. The matching of the templates designs two kinds of solutions to solve it. Dynamic time warping technique and vector quantization techniques are used in detecting real-time data. Besides, this thesis tests their feasibility by a large number of tests trials. The experimental result indicates that the system can detect given advertisement very well.