Abstract
The Generalized Cross Correlation (GCC) framework is one of the most widely used methods for Time Difference of Arrival (TDOA) estimation and Sound Source Localization (SSL). TDOA estimation using cross correlation without any pre-filtering of the received signals has a large amount of errors in real environments. Thus, several filters (weighting functions) have been proposed in the literature to improve the performance of TDOA estimation. These functions aim to mitigate TDOA estimation error in noisy and reverberant environments. Most of these methods consider the noise or reverberation and as one of them increases, TDOA estimation error increases. In this paper, we proposed a new weighting function. This function is a combined and modified version of Maximum Likelihood (ML) and PHAT-rg functions. We named our proposed function as Modified Maximum Likelihood with Coherence (MMLC). This function has merits of both the ML and PHAT-rg functions and can work properly in both noisy and reverberant environments. We evaluate our proposed weighting function using real and synthesized datasets. Simulation results shows that our proposed filter has better performance in terms of TDOA estimation error and anomalous estimations
Highlights
Sound Source Localization (SSL) has many applications in military and civilian areas such as mixed audio signals separation, robotics, video conferencing, speech enhancement, tracking of acoustic sources, underwater acoustics, and advanced hearing aids [1,2,3,4,5,6,7]
By comparing the mode values of di erent weighting functions in this table, we found that three weighting functions PHAT, Maximum Likelihood (MML) and Maximum Likelihood with Coherence (MMLC) have the nearest values to the Time Di erence Of Arrival (TDOA) of the main source (TDOA of the main source is {3.20), but by comparing the mean and mode values of these three functions, we can conclude that the MMLC function has the lowest di erence between mean and mode values
We proposed a new weighting function for the Generalized Cross Correlation (GCC) framework using a combination of modi ed Maximum Likelihood (ML) and PHAT- functions
Summary
Sound Source Localization (SSL) has many applications in military and civilian areas such as mixed audio signals separation, robotics, video conferencing, speech enhancement, tracking of acoustic sources, underwater acoustics, and advanced hearing aids [1,2,3,4,5,6,7]. The angle at which the power reaches its maximum value is the Direction Of Arrival (DOA) of the sound source These algorithms have good stability in direction estimation, but their computational costs are very high. High-resolution spectral estimation methods, which are famous in subspace approaches, use modern spatial- ltering methods and are used in narrowband and far- eld signal processing In speaker localization, these methods deal with constraints that limit their e ectiveness. These methods deal with constraints that limit their e ectiveness These algorithms are signi cantly less stable than the beamforming approaches due to source and microphones modeling errors. These errors are due to non-ideality in signal propagation, nonlinear properties of microphones, and variations of source position Like beamforming algorithms, these approaches are based on spatial search and have high computational cost [10].
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
