Extracting room geometry from impulse response measurements
MetadataShow full item record
- Institutt for fysikk 
An algorithm for establishing room parameters from measured impulse responses is created and implemented in matlab. The script is defined for synthetic and measured impulse responses. The setup is optional placement of a microphone cluster and source in a room. The necessary input for the algorithm is the pairwise distances between the receivers, and the impulse response received for each microphone. The algorithm is tested with synthetic impulse responses as well as real data experiments. The method for finding the walls is performed by first locating image sources from specular reflections in impulse responses, and then computing the plane equations.The distances between the microphones were measured using three different methods. First, a laser scanner was used for measuring the microphone positions in the room. Knowing their x-, y-, and z-positions, the pairwise distances between all microphones could be calculated. Secondly a digital caliper was held by hand over the middle, on the top of the microphones. The third method was that of using impulse responses and time of arrival for the direct sound for pairwise microphones.The difference in the microphones distances between impulse response measurements and laser measurement gave an average mean of 7.715 mm with a standard deviation of 7.259. Between impulse response measurements and caliper measurements, the difference in positions had a mean of 1.39 mm and a standard deviation of 1.41 mm.Results of the implementation showed that the optimal number of microphone were five, and microphone distribution should be within a fictive cube with side length 0.5 m. An increase in the distances between the microphones in a cluster, resulted in an increase in the computational time.Implementations of synthesized impulse responses in a shoebox room, gave an average error in image source location of 31.1 mm while for the measured impulse responses the average error was 296.9 mm. Despite the large error in image sources for measured impulse responses the results still provide a good approximation for location of the walls, but in relation to the measured source and image sources, not the theoretical.Generally, the algorithm shows good potential for establishing room parameters from measured impulse responses.