Content-Based Audio Retrieval Using Perceptual Hash

In this paper, a two-stage audio retrieval algorithm based on perceptual hash is proposed. In the first stage, the n-gram algorithm is used to index the perceptual hash values and filter the irrelevant candidates. In the second stage, the bit error rate (BER) is computed for exact searching. Experimental results demonstrate the effectiveness of the two-stage coarse-to-fine method with high precision and recall values.