D τ which corresponds to the TDOA between microphone 1 and 2 The TDOA estimate

# D τ which corresponds to the tdoa between microphone

• 122

This preview shows page 54 - 56 out of 122 pages.

D τ ), which corresponds to the TDOA between microphone 1 and 2. The TDOA estimate is the time lag that maximizes ) ( 12 τ R : ) ( max arg ˆ 12 12 τ τ τ R D = Note that finding 12 ˆ τ requires a simple, one-dimensional search. In general, Equation 4.9 has multiple local maxima. The amplitudes and corresponding time lags of these maxima depend on a number of factors. These factors include the separation distance of the microphones, the nature of the source signal and noise signals, and the choice of the weighting function ) ( 12 ω Ψ . 4.1.1 Maximum Likelihood (ML) Weighting Function When there is no multipath propagation (no reverberation), and the source and noise terms are uncorrelated Gaussian signals, the maximum likelihood (ML) weighting yields an estimator that is asymptotically unbiased and efficient. The ML weighting function is given in terms of the power spectral densities of the source signal, ) ( ω S , and noise signals, ) ( 1 ω V and ) ( 2 ω V , [64]: 3 This finite range is determined by the distance between the microphones divided by the speed of sound. 44
1 2 1 2 1 12 ) ( ) ( ) ( ) ( 1 ) ( ) ( ) ( ) ( + + Ψ ω ω ω ω ω ω ω ω V S V S V V S ML The idealized conditions for which this weighting is optimal are rarely encountered in practice. It has been shown that even mild reverberation greatly degrades performance of the ML estimator [19]. The coherence function, on which the ML estimator is based, is estimated in practice using a temporal averaging technique, such as the one described in [18]. However, this process can be problematic for non-stationary source signals, such as long segments of speech. An approximation to this weighting, which has been shown to work well with speech signals, operates on a single, short segment of speech, and can be given in terms of the magnitude spectra of the microphone signals and noise signals [11]: 2 1 2 2 2 2 2 1 2 1 12 ) ( ) ( ) ( ) ( ) ( ) ( ˆ ω ω ω ω ω ω X V X V X X ML + Ψ Here ) ( 1 ω X and ) ( 2 ω X are the received microphone spectra, and ) ( 1 ω V and ) ( 2 ω V represent the additive noise components that are assumed to be estimated over silence regions. When reverberation is present in the noise terms, the basic assumption that the noise and source signals are uncorrelated is violated. While ML-type weightings are widely used in speech-array applications, they are inadequate in reverberant environments and will not be used in the experiments of this thesis. 4.1.2 The Phase Transform (PHAT) Weighting Function Another weighting function, known as the phase transform (PHAT) [64], is sub-optimal under reverberation-free conditions, yet performs considerably better than ML in realistic environments. It is a popular form of GCC because of its robustness to reverberation. GCC-PHAT whitens the microphone signals, which in turn whitens the cross-spectrum, ) ( ) ( 2 1 ω ω X X . It is defined as follows: ) ( ) ( 1 ) ( 2 1 12 ω ω ω X X Ψ (4.10) GCC-PHAT has been shown to be effective in real environments [72][93]. It will be studied in more detail later in this thesis.

#### You've reached the end of your free preview.

Want to read all 122 pages?

• One '14

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern