This paper proposes a method to separate speaker information from phonetic information included in speech data. A speaker recognition method using the separated speaker information is also proposed and shown to be equivalent with a method based on speaker subspace. The validity of this proposed method was verified by earring out simple speaker recognition experiments.