Wake Word Based Room Identification with Personal Voice Assistants

Personal Voice Assistants (PVAs) are used to interact with digital environments and computer systems using speech. A wake word such as ’Alexa’ is spoken by the user to initiate interaction with the PVA. We use the audio recording of the wake word to determine the room in which user-PVA interaction takes place. We collected data from 10 different rooms in which a user speaks the wake word at different locations. This dataset is used to evaluate three different neural network based algorithms for room identification. Our evaluation shows that rooms can be identified with 90% accuracy. The impact is twofold: (i) PVA audio recordings leak private information about the user environment; (ii) Acoustic room identification is an option for augmenting user-PVA interaction.