The goal of the project is to construct a system for automatic extraction of audio files containing a tune performed with desired timbre, given microphone vocal input. The system will be able to identify which audio files within the searched database match the tune and timbre of specified instrument or instrument family with selected articulation. The approach consists of sound processing, segmentation, and indexing of audio files for content-based searching. The first step is the elaboration of the appropriate sound description for timbre classification, based on low-level sound parameters. The main process involves segmentation of audio files from the database into musical events, and indexing them using a dedicated tree-based structure. When this step is accomplished, the database of audio files, indexed using the above procedure, can be searched.
This research project has a strong educational component. It involves graduate students gaining expertise in sound processing under the guidance of PIs and adding this new research domain into a set of topics currently offered by the graduate program in Computer Science at UNC-Charlotte. Involvement of faculty and graduate students from the Polish-Japanese Institute of Information Technology in this project will add not only a strong international component to it, but also will expose graduate students at UNC-Charlotte to international collaboration.
As the result of this project, new tools applicable for automatic search of audio files matching desired contents, including melody and timbre information, will be available to users and research community.
