Improving Spatial Sound Capturing on Mobile Devices Through Fusion of Inertial Measurement Data

Sammanfattning: Through the use of sensor arrays it is possible to extract spatial information about signals present in the environment. For instance position, velocity, distance, etc. In this thesis we focus on the use of microphone arrays with the aim of accurately determining, and over time track, the Direction of Arrival (DoA) of nearby sound signals impinging on the array. As modern mobile devices like smartphones and tablets are commonly equipped with two or more microphones, these constitute a simple microphone array. This gives us the opportunity of attaining the above aim. However, in the scenario that the microphone array is not stationary, a number of problems arise. In this thesis we implement an algorithm to estimate the microphone array's orientation in three-dimensional space, with the aim of using these estimates to cancel the effect of the array's orientation on the DoA estimates. The cancellation is done with a dynamical model, which we use in a modified Kalman Filter (KF) capable of tracking an arbitrary number of sound sources over time. Additionally, we estimate the computational cost of the mentioned algorithms. The simulation results show satisfactory performance in the modified KF with respect to handling crossing trajectories and noise in the measurements. The chosen algorithm for orientation estimation proves susceptible to magnetic disturbances to the extent that usage in the context of mobile devices is undesirable. Due to this, the orientation estimates are provided by a proprietary algorithm. Experiments where the DoA of the sound sources are computed using actual sound, together with orientation estimates, confirm the correctness of the proposed model.