Results

Below we provide the full result tables from our experiments that test the BarrierBypass attack.
(Note: In the paper we include condensed versions of these same tables)
We also provide some new results from additional experiments that were conducted to test louder command SPL levels and farther VA device distances (not in the current paper submission).

Full Result Tables

TABLE 1: Wake Word injection success rates, for attacking the Amazon Echo Dot 2, for each Barrier scenario.

TABLE 2: Wake Word injection success rates, for attacking the Google Home mini, for each Barrier scenario.

TABLE 3: Command injection success rates, for attacking the Amazon Echo Dot 2, for each Barrier scenario.

TABLE 4: Command injection success rates, for attacking the Google Home mini, for each Barrier scenario.

Additional Experiments (not currently reported in the submission)

TABLE 5: New experimental results for the across-window and across-wall attack scenarios when testing louder command SPL levels (95, 105 dB) and farther VA device distances (0.5, 1, 2 meters).

Discussion: From the results above, we can see that the threat level of the BarrierBypass attack increases when the command SPL level is increased, achieving successful attacks at farther distances. This demonstrates the feasibility of this attack as there are many scenarios where loud command SPL may be used. And from our demo videos, we can see how the loudness of the commands is severely decreased after passing a barrier so there is potential for the audio to go unnoticed by a user that is inside.

Page updated

Google Sites

Report abuse