In the paper, we describe a system SIRIUS for recognition of continuous Russian speech, which is developed in the group of speech informatics of SPIIRAS. The specific feature of this system is that the language and speech are represented on morphemic level. This allows one to significantly reduce the size of lexically recognizable dictionary and increase the processing rate. We describe the process of introduction of the Russian speech recognition system into the area of infotelecommunications for voice access to the Internet-version of the electronic catalogue “Yellow Pages of Saint Petersburg” with the purpose of creation of an automated call-center for answering subscriber’s calls. In the paper, we demonstrate the results of testing the system work with speech samples recorded both in offices and in conditions of phone conversations.
展开▼