Personalized speech recognition for Internet of Things

With the rapid increasing of smart devices, there has been a growing interest in the concept of Internet of Things (IoT). While as a network of connected objects, IoT is created by enabling machine to machine interactions, another important factor of IoT is the human-machine interaction. By creating a connected life, people can interact with their devices, appliances, vehicles, etc. As one of the most natural ways of communication, using speech to interact with things adds value by enriching the user experience. As a subset of internet of things, smart/connected home is a fast growing industry that includes automation of household activities (e.g., centralized control of lighting, heating, air conditioning, etc.) for more convenience, as well as added security (e.g., control of alarm system, doors, windows, etc.). This work focuses on a combination of speech recognition and natural language understanding for smart home applications with customizable devices. Dynamic hierarchical language models are discussed and shown to be effective in creating personalized speech recognition, and evaluated in terms of both word accuracy and semantic accuracy.