Implementation and performance issues from a voice activatednetwork control trial

摘要

Roll out of voice recognition-based interfaces in telephonyapplications continues to grow as the implementation and deployment ofspeech processing technologies become more feasible and robust. Oneparticular voice recognition application that was trialed by US WEST wasa voice activated network control (VANC) service based onspeaker-independent continuous-speech recognition (SIR) technology. TheVANC service allowed users to control custom calling services (CCS) andcustom local area signaling service (CLASS) type network features withvoice commands in lieu of DTMF (dual tone multifrequency) codes. Aflexible and scalable system was designed to support residentialtelephone customers. In this paper, a number of key functionalityrequirements, usage characteristics, and recognition performance issuesconcerning the implementation and use of the system are discussed