A method includes detecting an initial vocalization by a user in an environment including a primary device and a secondary device, analyzing the initial vocalization to determine whether a future vocalization is likely to contain sensitive information based on the initial vocalization where the future vocalization is a potential future vocalization, and controlling, based on determining that the future vocalization is likely to contain sensitive information, a speaker of the primary device to output a noise canceling signal configured to prevent the secondary device from detecting the future vocalization.