Selection of melody lines for music databases

摘要: One major approach to music retrieval is model as a sequence of features, after which traditional information techniques are applied on the sequence. Because temporal nature and inexactness user queries, most effort systems focus issues such indexing approximation match. In contrast, processing before feature extraction, identification melody track, were often considered easy or done. This may be case in controlled environment, one for musicology research, where pieces carefully analyzed by human beings being submitted database. However, an environment large volumes obtained from Web, manual analysis impractical. Since many well-known musical features pertain pieces, users remember song, algorithms that select tracks piece important Web-based content-based systems. We describe number automatic track selection context. also study performance comparing their answers those judged beings.