A dual compression voice recordation non-repudiation system provides a voice recognition system that compresses voice samples for two purposes: voice recognition and human communication. The user navigates through menus displayed to a user through a television set using both voice commands and button presses on a remote control. The invention accepts voice samples from the remote control and compresses the voice sample for voice recognition while a copy of the voice sample is stored on a storage device. Compressed voice samples are sent to a Voice Engine that performs voice recognition on the voice sample to verify if it is from an authorized user from the consumer's household and determines the action required. If the command is to form a contractual agreement or make a purchase, the Voice Engine determines the merchant server that is appropriate for the action and sends the action request to the server. The voice sample is compressed for human communication and stored on a storage device along with any additional information. The stored human communication compressed sample and any additional information on the storage device may be later retrieved and the human communication compressed sample decompressed into a form that can be played back when a user attempts to repudiate a contractual agreement or purchase. Alternatively, the invention performs both compressions at the same time.