Speech recognition engine adaptions for smart home dialogues

This paper considers the needs of speech recognition-based dialogues for smart homes, and proposes a structure to allow effective speech recognition in such circumstances. A smart home system based around the Freevo home theatre platform, and the Sphinx2 speech recognition engine has been designed to implement the features and grammar optimization described. In addition, the particular requirements of complexity and size minimisation for embedded systems are discussed. In practical terms we propose a method of continuously variable vocabulary size to maintain required speech recognition accuracy in a smart home context.