An environmental condition detector emits only safe, validated verbal messages recorded in a user's voice to help occupants of a building respond safely to sensed environmental conditions. The message emitted in the user's voice is generally a familiar voice to other building occupants to better capture their attention, and only certain, safe, user messages are validated for use thereby eliminating the chance that the user may unknowingly record an unsafe message to be emitted during a sensed environmental condition. The verbal message is recorded by a user and is analyzed through speech recognition to determine if the user-recorded message can be validated as containing safe content. A user-recorded message found to match manufacturer preprogrammed message content is validated and selected to be audibly emitted in the user's voice when an environmental condition is sensed. Various embodiments include single station units, interconnected units, or centrally controlled security systems.