This paper describes a speaker-adaptive word recognition system in which word templates are indirectly atuned to a new speaker through an adaptation of the code book for vector quantization. The adaptation of the code book is made by the following four steps.