Visualization tools and graphical methods for source localization and signal separation
Gaute Havik Eriksen · NORA - Norwegian Open Research Archives · 2006
Much effort has been used on the problem of blind source separation (BSS). The problem is the one of separating audio sources present in a room, using a set of differently placed microphones capturing the auditory scene. An estimate of the number and position of sources in the recorded sound, would simplify the sound separation process. We propose two different methods, the Cross-correlogram (CCG) and the Delayogram (DG), for visualizing of sound. From the produced images from the CCG and the DG, the number of sources in the recordings can be found. The images are important in the source positioning process, where we introduce ``Signature matching'', a novel method utilizing the ``Microphone-fan''(M-fan) for source positioning. The M-fan is a new device for sound recording with two rotating microphones. Though movement normally is treated as a disadvantage in source separating, we show that the movement can be used constructively for source position estimating. From the images produced by the CCG and the DG, the signature matching process estimates in which direction the source is located. By combining two M-fans, an estimate of the spatial position of the sources is obtained. In addition, the signature matching process shows strong robustness against noise. The results are validated with simulated test cases, and illustrated with real environment recordings. Besides being used for preprocessing in the BSS problem, our proposed methods can be utilized in e.g. mobile communication, person and equipment tracking, automatic camera and microphone guiding systems, etc.