This paper presents a public database of anechoic audio and 3D-video recordings of several small music ensemble performances. Musical extracts range from baroque to jazz music. This work aims at extending the already available database of anechoic stimuli, providing the community with flexible audio-visual content for virtual acoustic simulations. For each piece of music, musicians were first close-mic recorded together to provide an audio performance reference. This recording was followed by individual instrument retake recordings, while listening to the reference recording, to achieve the best audio quality. In parallel, 3D-video content was recorded of each musician, employing three Kinect 2 RGB-Depth sensors, allowing for the generation and easy manipulation of 3D point-clouds. The position/orientation of moving instruments extracted from these videos allows for the incorporate dynamic directivity in the auralizations. This article details the choice of the musical pieces, the recording procedure, and technical details on the system architecture including post-processing treatments to render the stimuli in immersive audio-visual environments.