The development of computational tools for musicological studies and musical analysis has been a very active area of research in recent years. Some of these tools operate on symbolic representations of music, while others process the audio from a recorded performance. In both cases, an annotated dataset representative of the corpus under study is essential for the research. This paper reports the generation of a database of audio and video recordings of Candombe performances, including detailed descriptions of the process of curation and production, technical aspects, musical content, and annotations. It is part of an ongoing interdisciplinary research effort for the development of tools for computer-aided music analysis. This encompasses several research problems concerning the development of technologies capable of extracting musically meaningful information from a recorded performance. A significant part of the recordings and annotations is being released for the research community. Of course, apart from its research purposes, the database has also a documentary value.