PURPOSE: To improve a recognition rate by varying the recognition candidate number in accordance with the SN ratio, which varies due to environmental conditions, in the recognition candidate number setting system of a speech recognition device which outputs plural recognition candidates by computing the degree of similarity between the plural voices registered beforehand and the inputted voices.

CONSTITUTION: Plural voices are previously registered in a dictionary 12, the inputted voices are analyzed in an analyzing section 11, a score computing section 13 computes the degree of similarity between the result of the analysis and the voices registered in the dictionary 12 and a recognition candidate decision section 14 decides the recognition candidates which are selected from the higher degree of similarity to the prescribed order and outputs the recognition candidates. In this voice recognition device, an SN ratio computing section 15, which computes the ratio of the inputted voices to the environmental noise (SN ratio), and a recognition candidate number controlling section 16 which controls the number of the previously mentioned recognition candidates are provided.