This paper propose a based CASIT [5] audio retrieval model. CASIT method is developed for our text information retrieval system witch gives good results. We have extended the CASIT method to the video and audio applications. In this paper we present its application in the audio content-based retrieval. The adapted method is called CASIA (in French: CAlcul de la SImilarité Audio). CASIA identify subsequence of audio using basic audio descriptors. Each audio document is pre-processed before the extraction of a set of basic audio descriptors, which characterize the temporal and the spectral information. Therefore an audio signal is represented by a characteristic sequence called in this paper "sequence fingerprint".