An Adaptive Algorithm for Speaker Localization in Real Environments using Smartphones

The direction of arrival (DoA) of a sound source is commonly computed using time delay estimation (TDE) algorithms. In this work, an adaptive eigenvalue decomposition (AED) algorithm is integrated with a noise reduction method to handle speaker localization in real environment conditions. Conditions involving low signal-to-noise ratio and high reverb time affects the reliability of localization systems. Results yielded an on-point accuracy of 76.56% which is 9.79% greater than a widely used cross-correlation algorithm for TDE. The system also obtained 86.63% region accuracy on a smartphone simulation experiment.