Skip to Main Content
A number of content-based music retrieval systems have been presented in the last few years. Although query-by-humming has been one of the preferred strategies, the translation of voice input into note-like attributes has been tackled with naive algorithms. In this paper, we present a system developed to consider the peculiar facets of the singing voice. A pre-processing block for signal enhancement, an algorithm for pitch detection and a labeling stage for fine adjustment of intonation are addressed in the context of querying by singing. Experimental results confirm that this system is able to translate a sung performance into a representation close to a music score, thus particularly suited for melody-based search.