Sound localisation

Abstract

This dissertation aims to implement a solution for sound localisation by trading off between cost and result accuracy, by using industry-standard equipment and setup. The solution must provide a real-time result: the code behind the implementation must be light enough to be able to keep up with this requirement. The system consist of four microphones which are pre-amped, and fed through a DAC, where MATLAB analyzes the sound windows captured and performs tests based on the time delay between the signals in microphone pairs known as ITD (Interaural Time Delay) or TDOA (Time Difference On Arrival). This data is interpreted using two different hardware topologies and two different analysis methods. Tests conclude that the system works best when the distance between microphones 1 and 2, and that between microphones 3 and 4 is 0.16m, and the distance between microphones 2 and 3 equates to 0.95m. In most cases the system detects direction far more accurately than distance. These anomalies are due to a number of external factors which the system does not take into account such as the presence of other background noises which attenuate the prominence of the required signal, together with any reverberations present in the room. This analysis, performed using a Cartesian translation of real-life environment, can be performed by the system in 2.107s. Provided that the sound source is not in quick motion, this processing time is quick enough to pinpoint the sound source’s location.