Thumbnail Image

MIRACLE - Microphone Array Impulse Response Dataset for Acoustic Learning

Kujawski, Adam; Pelling, Art J. R.; Sarradj, Ennes

This work presents a large dataset of measured impulse responses of spatially distributed sources in a plane parallel to a planar microphone array in an anechoic environment. It can be used for various acoustic machine learning tasks and as a benchmark for data-driven modelling and interpolation methods. The dataset, which we call MIRACLE, contains a total of 856, 128 single-channel impulse responses across four different measurement scenarios. A regular grid of 64 × 64 source locations was sampled for two different source plane to array distances. The dataset also contains measurements on a densely sampled 33 × 33 grid for the short distance, as well as measurements with the presence of a reflective panel for the longer distance. We assess the quality of the provided source location labels and provide directivity measurements of the loudspeaker used for excitation.