SPEAKER RECOGNITION DEVICE, SPEAKER RECOGNITION METHOD AND PROGRAM

PROBLEM TO BE SOLVED: To simply and easily perform speaker recognition. SOLUTION: A prediction coefficient for predicting a speech waveform is stored for each speaker in a coefficient storing section 40. A prediction waveform creation section 33 performs arithmetic using a speech data of an input speech and the prediction coefficient, and generates a prediction waveform for each speaker. A prediction error calculating section 34 calculates an error of the prediction waveform to the speech waveform of the input speech for each speaker. A speaker specifying section 35 specifies the speaker of the input speech based on the calculated error for each speaker. As for a prediction coefficient, a linear prediction coefficient which is obtained by performing linear prediction analysis on the speech data of the speaker is used. Processing such as converting a time domain to a frequency domain is not needed to perform, and speaker recognition is performed in the time domain. COPYRIGHT: (C)2010,JPO&INPIT