A feature extractor extracts feature quantities from a digitized speech signal and outputs the feature quantities to a likelihood calculator. A distance determiner determines the distance between a user providing speech and a speech input unit. The likelihood calculator selects registered expressions for speech recognition from a recognition target table based on the determined distance, to be used in calculation of likelihoods at the likelihood calculator. The likelihood calculator calculates likelihoods for the selected registered expressions based on the feature quantities extracted by the feature extractor, and outputs one of the registered expressions having the maximum likelihood as a result of speech recognition.