SMILE: a System for Content-based Musical Information Retrieval Environments

This paper reports on a system for musical information retrieval environments called SMILE. SMILE implements a methodology designed to cope with content-based musical retrieval. The innovation provided by the methodology is due to the synergic combination of sound information retrieval techniques and computer-based melodic surface extraction algorithms. The innovative corollary is that the latter extracts content-descriptive musical phrases from both documents and queries by thus permitting a clear separation between information being searched and indexes describing the semantic content. The information retrieval techniques are employed to index, retrieve, and rank by score the musical documents judged relevant to the final users' information need. Two types of retrieval functions are provided -- a querying function based on a virtual keyboard played by the user, and a browsing function to navigate an automatically constructed hyper-music which is made of documents and phrases. Hyper-music links are constructed automatically and are based on content, i. e. the algorithms used to detect links employ the index-based description of the musical document content.