Deciphering the Secrets of Musical Intelligence
As part of their trade, musicians continuously study and analyze music scores. Important musical tasks such as selecting repertoire require searching scores for their constituent elements. Searching collections of scores can be burdensome and cognitively taxing even for expert musicologists. Despite the promise of computing to facilitate searching over vast volumes of information, commercial search engines only operate on score metadata (e.g., composer, time period, genre, etc.). In this project, we have explored the user requirements, design choices, and software architecture of a search engine for querying music scores beyond metadata (e.g., instrument range, key/time signature, dynamics, etc.). We have also developed our proof-of-concept implementation of this architecture Ask Toscanini! which supports a wide selection of user queries, expressed as structured text strings, against a collection of digital scores. The engine can
search through any collection of scores, provided in the popular standardized MusicXML format, which is subsequently transformed into our custom search-efficient format. In addition to ensuring search efficiency, our design also renders the engine amenable for use by musicians, the majority of whom are not computing experts. The insights derived from this project can help future research efforts in enhancing search technologies for music scores and can serve as a blueprint for creating commercial solutions in this domain.
ICAT 2017-2018 Major Sead Grant