A Perspective on Speech Recognition

Communication between humans and machines by natural spoken language is a difficult but worthy goal toward which real progress has been achieved in the past thirty years. Although the ultimate goal of automic speech understanding at human levels of performance still eludes us, speech recognition technology has been useful for specific well designed applications since the late 1970's. Today, commercially available hardware exists for continuous speaker independent recognition of small vocabularies (e.g. digits) over the telephone network, as well as speaker dependent recognition of vocabularies of several thousand isolated words.