In a voice authentication apparatus, a characteristics analyzer analyzes characteristics of a sample noise which is generated around a subject while the subject generates a sample voice for authentication of the subject. A setting part sets a correction value according to the characteristics of the sample noise analyzed by the characteristics analyzer. A correction part corrects an index value, which indicates a degree of similarity between a feature quantity of a reference voice which has been previously registered and a feature quantity of the sample voice obtained from the subject, based on the set correction value. A determinator determines authenticity of the subject by comparing the corrected index value with a predetermined threshold value.