Speech interface for operating information technological home appliances

Information technological (IT) home appliances are becoming highly advanced by being connected to the Internet. Conventional push-button systems on remote controllers cannot suffice for IT home appliances because of their complicated requirements for versatile and complex functions. Speech input is expected to be a good substitute or supplement for push-button systems. Effects of introducing speech input to a TV system are evaluated, implementing facilities of on-line program retrieval besides ordinary functions of controlling the TV sets. One of the problems with introducing speech input is degradation of speech recognition rate due to being affected by surrounding noises and waves reflected on the surface of walls, ceiling, and so on. So, the test system introduced reduction of known noise, dereverberation scheme, delay-and-sum beamformer and blind source separation by independent component analysis as preprocessing for speech recognition. Effects of introducing these types of preprocessing are investigated comparing speech recognition rates in case of employing these types of preprocessing with those in the case of not employing these types of preprocessing. [Work supported by Knowledge Cluster Project, MEXT, and by Academic Frontier Project, Doshisha Univ.]