Probabilistic approach for building auditory maps with a mobile microphone array

Mapping sound emitting structures in 3D

“…But the performance of the method is somehow robust to the choice of these parameters as the worst result for the large aperture with a range of 4 m is no catastrophic considering the size of the grills. As a comparison, the maps presented in [11], [10] also exhibit localization errors in the 0.2∼0.3 m range for the 2D case. Fig.13 illustrates the effect of the log-odds limitation L max (the lower limit is set to L min = −L max ) in the case of the small aperture with a range of R max = 3.…”

“…Namely a ray is casted from the point {x r (t), y r (t), z a } in the direction {θ r (t) + θ i , φ i }. The rationale behind the use of ray casting is to trace back the sound until its sources as in [10]. However, by using a 3D representation of the environment, the sources are not restricted to be in a plane as in [10].…”

“…From the sound source localization literature and the idea behind the SRP approach, it is expected that a large power should correspond to a strong belief. For example in [12], [10] a scale version of the power was used as likelihood. In audio source tracking, creating a likelihood by scaling the power is also common [18].…”

“…As illustrated by this example, the created sound map can be a powerful condition monitoring tool for detecting structures emitting sound. Moreover, it is important to stress that for detection of sound sources in real structures, the proposed 3D sound map approach is an improvement over the maps obtained with the methods in [11], [10].…”

“…In [10], we proposed a method for building probabilistic maps of sound sources that reduces the complexity of the grid approach in [11] by considering only grid cells that are occupied by an object as a potential sound source candidate.…”