The audio reference pattern study device, the record media which records manner and the audio reference pattern study program

摘要

To provide a voice standard pattern generating means which has a small decrease in recognizing performance even when training data for generating a standard pattern for recognizing a voice passed through some line A are not much as to a procedure of learning for generating the voice standard pattern used for voice recognition. SOLUTION: A training data conversion means 105 converts training data of a line B stored in a B2 training data storage means 106 by using conversion parameters stored in a conversion parameter storage means 104 so that they approximates the data distribution of a line A and passes the data to a standard pattern generating means 108. The standard pattern generating means 108 outputs a voice standard pattern by using the training data of the line B converted to a distribution close to the line A by the training data converting means 105 and existent training data of the line A in combination.