Implementation and performance issues from a voice activated network control trial

摘要

Roll out of voice recognition-based interfaces in telephony applications continues to grow as the implementation and deployment of speech processing technologies become more feasible and robust. One particular voice recognition application that was trialed by US WEST was a voice activated network control (VANC) service based on speaker-independent continuous-speech recognition (SIR) technology. The VANC service allowed users to control custom calling services (CCS) and custom local area signaling service (CLASS) type network features with voice commands in lieu of DTMF (dual tone multifrequency) codes. A flexible and scalable system was designed to support residential telephone customers. In this paper, a number of key functionality requirements, usage characteristics, and recognition performance issues concerning the implementation and use of the system are discussed.