Human echolocation inside buildings

Classically, the term "sound recognition" is associated with understanding of speech, or recognition of a sound-producing source. However, information can also be acoustically extracted from a passive environment, namely by so-called "echo-location" or "echo-recognition." In the latter case, an observer is producing a sound that is traveling through the environment, and reflecting on different objects. By interpreting the reflected sound, information on reflectors can be retrieved. Echo-location is mainly being used by visually impaired people making click or hiss sounds and by some animals emitting ultrasonic bursts. Till now, most studies have been focusing on determining the location or size of a reflector. In this work, we have gone a step further and investigated to what extent aspects of the shape or texture of a reflector can be distinguished when comparing echoes of reflectors with different morphologies. The study was carried out by simulating the acoustic wave propagation from an observer producing sound to a reflector and back. Simulations for different scenarios were auralized and presented to listening subjects through headphones. The listening tests aimed at determining the degree of discrimination ability people have when pairwise comparing walls with different textures. The question on which temporal and/or spectral features could be used most efficiently as a cue for successful discrimination was addressed by means of artificial neural network recognition. The difference recognition performance for different wall texture pairs between the neural network and the average of a group of test persons was used to get insight in the cues that are used by people. Further insight in the echolocation process was obtained by artificially modifying features of the received sound and verifying the effect on the discrimination performance. The study was complemented by a listening test campaign in which the ability to distinguish filtered noise signals with a dip or peak in their spectrum was assessed, for different dip and peak widths and contrast levels. Both fixed pair comparison based and adaptive listening tests were used.