To perform voice recognition more exactly in a momentarily changing environment.

The on-vehicle voice recognition apparatus extracts a non-utterance part in acquired voice and stores it as background voice with priority. The apparatus recognizes the utterance part of the voice using a voice model and the two or more background voices extracted in descending order of priority. The apparatus outputs a candidate for words with highest reliability as the final voice recognition result. When the apparatus accepts a request for correction of the recognition result from a user, the priority of the background voice used for recognizing the words by high score is lowered.