Detecting musical solos in recorded concerts to extract individual musical performances in the Montreux Jazz Festival digital archive.
In order to enrich the archive, it is useful to identify when instrumental solos take place in recorded concerts. Musical solos include; vocals, saxophone, bass, guitar and drums.
As a result, a tool was designed at LTS5 to automatically detect all solos in the Montreux Jazz Festival digital archive at 85% accuracy. It enables the referencing of all musical solos stored in the archive for easy search and retrieval of solos from favorite bands. A second solution is currently being developed at IDIAP.
The input audio file is first cut into frames, from which characteristic features are computed. These features are then processed by a classification algorithm which was previously designed to recognize solo from non-solo parts. Finally the algorithm automatically decides for each second of the input audio whether it is a solo performance or not.
LTS5 (Signal Processing Laboratory)