Feature extraction apparatus and method and pattern recognition apparatus and method

A pattern recognition method for recognizing a pattern of time serial unclassified input data by classifying the unclassified input data as one of a predetermined number of classes, comprising:

• a framing step of sequentially extracting parts a plurality of frames from the unclassified input data at predetermined frame intervals and outputting each extracted part frame as frame data;
• a calculating step of calculating an average vector and a variance matrix as a feature distribution parameter including an average vector and a variance matrix for each frame of the plurality of frames extracted from unclassified input data, said feature distribution parameter representing a distribution obtained when the frame data is mapped to a space of a feature quantity of the unclassified input data;
• and
• a classifying step of classifying a series of the feature distribution parameters as one of the predetermined number of classes,
• wherein the classifying step includes a judging step of judging, by using one hidden Markov model method, which of the predetermined number of classes the series of feature distribution parameters belongs to.