Unsupervised Improved MVDR Beamforming for Sound Enhancement

<p>MCFSTD was created with the aim of evualuating multi-channel sound separation and localization. It is introduced in our article titled "Unsupervised Improved MVDR Beamforming for Sound Enhancement". It was recorded in 4 rooms with 3 microphone arrays. The zip files are organized by rooms. The "home" room is an appartment living room, the "2008" room is a conference/meeting room, the "1002" room is a dining, and the "1004" room is a very large room. The recording positions are described in the article.</p>\n<p>The radius of the circle on which was placed the loudspeaker differed from room to room depending on what the size of the room allowed. The dining hall had a 2.5 meters radius, the large room had a 3 meters radius while the living room and the conference room had a 1.5 meters radius. The base of the loudspeaker is at 86 cm from the ground and the top is at 114 cm. As there are 3 microphone arrays recording at the same time, they could not be placed exactly at the centre of the circle. The origin of the coordinates system used for the following positions is placed on the ground at the centre of the circle which the loudspeaker plays the audio recordings. The x-axis goes towards the G position and the y-axis towards the A position. The 16Sounds USB microphone array was placed at 26 cm in x, 0 cm in y and 40.5 cm in z. The Kinect was placed at -11 cm in x, -26 cm in y and 76 cm in z. The ReSpeaker was placed at -20 cm in x, 12 cm in y and 63.5 cm in z. Moreover, the 16 microphones of the 16Sounds USB sound card were arranged in a rectangular configuration where the length is on the x-axis. The figure below shows this arrangement.</p>\n<p></p>