Audio segment retrieval using a short duration example query

We propose a general approach to audio segment retrieval using a synthesized HMM. The approach allows a user to query audio data by an example audio segment of a short duration and find similar segments. The basic idea of our approach is to first train a theme HMM using the given example and a general background HMM using all the audio data, and then combine these individual HMMs to form a synthesized "background-theme-background" HMM. This synthesized HMM can then be applied to any audio stream as a parser to detect the most likely theme segment. We overcome the problem of a short duration being used to train a theme HMM, by using the MAP rule with the background model as a prior model. Evaluation of the proposed retrieval scheme, using short duration example audio clips of narration as queries, gives quite promising results.