Simulation with Data

We conducted user testing to validate our simulation of moving sound using experimental data. We played four different audio clips for each user comprising of two different sounds (walking in grass or plain sinusoidal tone) with and without the Hamming window implemented. The user drew the trajectory they guessed the sound was following, and we compared their estimation to the actual trajectory, which was a circle for each audio clip. We had intended to create complex trajectories for our moving sounds, but realized during some preliminary testing with a few users and ourselves that even a simple circular trajectory was hard to detect accurately. As a result, we decided to keep our testing simple, and we use the circular trajectory depicted in Figure 9. The simulation starts by generating audio at zero degrees, which is directly in front of the user, and moves it five degrees at a time counter-clockwise until it gets back to the starting point, zero degrees.

We collected data from three users. Only three out of 12 estimated trajectories at all resembled the actual trajectory, and even then, it was only a vague resemblance to a counter-clockwise circle. One interesting finding from user testing was that while none of the trajectories in the four audio clips were accurately approximated, there did seem to be a difference between the estimated trajectories that used the Hamming windows and those that were not. The Hamming window clips had smoother trajectories -- there was less back and forth between the left and right ears. Figure 10 shows one user's estimated trajectories. We drew the blue arrows to indicate the direction in which they drew the lines. The Hamming-window audio clips are smoother in Figure 10 than the non-Hamming-window audio clips.

There are two key takeaways from our user testing: One, our simulation with experimental data is neither accurate nor consistent, and two, while using the Hamming window does not significantly improve the accuracy of sound generation, it does smoothen a moving sound's trajectory from the listener's perspective.

Also, most of our users said they found the walking clips to be "creepy", particularly the first one, which uses the Hamming window. They could easily tell it was not the original audio, but they still found it life-like enough to be creepy. So, that was one small success!