The results may seem to contradict prior results in text document modeling that have shown the advantages of latent topic model over latent semantic analysis [9, 10]. We argue that this is because only small number of words are available for each description of an audio clip, while previous reports were performed on text documents which include many words for a single document. As given in Table 2, the average number of words in a description is 7.2 which might be too small to train the topic models which utilize a probabilistic approach; LSA utilizes a deterministic method.