Source localization results by SRP and TF on the composite sequence in S 1 with

# Source localization results by srp and tf on the

• 6

This preview shows page 4 - 5 out of 6 pages.

Source localization results by SRP and TF on the composite sequence in S 1 , with W = 1 s. (a) SNR of the noisy input signal. (b)- (d) The confidence map of the clean speech, the noisy signal processed by SRP, and the noisy signal process by TF, respectively. (e)-(f) The confidence function in the 86-th block (43 s) for the clean speech and the noisy signal, respectively. camera. We measure the tracking error, bounded at 180 , at the video frame rate (30 Hz), and calculate the mean and standard deviation across the whole trajectory. We compare particle filtering (PF) with, as baseline methods, median filtering (MF) and no filtering (NF) on the localization at individual blocks. MF updates the localization at the b -th block as the median value, M ( . ) , of the localization results across a sequence of W p blocks: ˆ θ b MF = M ˜ θ b - W p +1 , · · · , ˜ θ b , (10) where W p is predefined constant. NF uses the localization at the b -th block without any processing: ˆ θ b NF = ˜ θ b . (11) We use four block sizes, W ∈ { 0 . 5 , 1 , 2 , 3 } s, and in each block we use a STFT of size 1024 and 50% overlap. We set the search area as [ - 180 , 180 ] with an interval of 1 , i.e. D = 361 . We set the noiseless sector as [ - 45 , 45 ] . For the particle filter we set N = 1000 , and we use a different set of parameters, empirically chosen, for each block size: for W = 0 . 5 s, σ p = 3 . 5 , σ u = 10 , and W p = 8 ; for W = 1 s, σ p = 5 . 5 , σ u = 4 . 5 , and W p = 4 ; for W = 2 s, σ p = 7 . 5 , σ u = 3 . 5 , and W p = 2 ; for W = 3 s, σ p = 9 . 5 , σ u = 3 and W p = 1 . σ ˙ p = 0 . 05 for all setups. Unless otherwise specified, W = 2 s in the comparisons. Fig. 5 compares the localization results obtained by the time-frequency (TF) and the steered response power (SRP) approaches [23], with W = 1 s, in the composite sequence recorded in S 1 . Fig. 5(a) depicts the temporal variation of the SNR, computed per processing block [30]. The SNR, which varies significantly across the blocks, is lower than 0 25 50 75 100 125 150 175 Time [s] -180 -90 0 90 180 DOA [deg] (a) 0 25 50 75 100 125 150 175 Time [s] -180 -90 0 90 180 DOA [deg] (b) 0 25 50 75 100 125 150 175 Time [s] -60 -30 0 30 60 DOA [deg] (c) Groundtruth (audio) Groundtruth (video) No filtering Median filter Particle filter 0 25 50 75 100 125 150 175 Time [s] 0 30 60 DOA error [deg] (d) Fig. 6. Tracking results on the composite sequence in S 1 , with W = 1 s. (a) Original peak detection result. (b) Proposed peak detection result. (c) Trajectories generated by different trackers. (d) Tracking errors obtained by different trackers. -10 dB in most blocks and can be lower than -25 dB in some blocks (e.g. between 30 s and 60 s). Fig. 5(b)-(d) shows the confidence map computed using clean speech, SRP and TF, respectively. In Fig. 5(b), the trajectory of the clean speech can be observed clearly. In Fig. 5(c), the trajectory of the ego-noise, but not that of the speech, can be observed. In Fig. 5(d), the trajectories of both the ego-noise and the speech can the observed. Fig. 5(e) depicts the confidence function computed by SRP and TF, respectively, at the 86-th block (around 43 s), where both approaches can detect the peak location correctly. Fig. 5(f) shows the confidence function computed by SRP and TF for the noisy signal in the same block. In this low-SNR scenario (-24.9 dB), SRP detects two

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

Want to read all 6 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

Stuck? We have tutors online 24/7 who can help you get unstuck.
Ask Expert Tutors You can ask You can ask ( soon) You can ask (will expire )
Answers in as fast as 15 minutes
Ask Expert Tutors