As users often produce new words which are "unknown" to the system, processing unknown words is a key issue to speech recognition systems in the real world. This paper describes an approach to the problem of detecting unknown words, especially in a large vocabulary speech recognition. The approach consists of utilizing the recognition scores obtained both with and without a pre-defined dictionary. We studied the relationship between vocabulary sizes and detection rates of unknown words using this technique. It is shown that the detection task of unknown words is comparable to the recognition task for vocabulary sizes of over 1000 words.
Keywords: Unknown Words, Speech Recognition
Bibliographic reference. Hayamizu, Satoru / Itou, Katunobu / Tanaka, Kazuyo (1993): "Detection of unknown words in large vocabulary speech recognition", In EUROSPEECH'93, 2113-2116.