The goal of this project is to develop new approaches to retrieve specific songs based on text queries, thus enabling users of the Museeka Music Information Retrieval (MIR) system to search for songs relevant to a textual query (e.g., as currently possible for textual files with search engines like Google). The main challenge in such a task is the semantic gap between, on one hand, the words used by the users to express what they would like to hear and, on the other hand, the musical characteristics (timbre, rhythm, etc.) automatically extracted from a song, complemented by additional information usually available from the web, such as performers, lyrics, etc. Bridging, at least partially, this semantic gap would significantly improve the effectiveness, hence the value, of the Museeka MIR system.