摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognizing method which does not require the changeover of a voice recognition mode for recognizing the voice of an unspecified speaker and the voice of a specified speaker. SOLUTION: An HMM state group without transition probability is registered in an unspecified speaker voice recognizing voice dictionary. When a specified speaker pattern is registered, voice inputted by the specified speaker is analyzed to generate a voice analysis string concerning inputted voice. The HMM state string being most approximate to the generated voice analysis string is obtained by retrieving the unspecified speaker voice recognizing voice dictionary and registered in a specified speaker pattern register memory as a specified speaker pattern. When the voice is recognized, inputted voice is analyzed to generate the voice analysis string for the input voice and the unspecified speaker pattern is compared with the specified speaker pattern which is registered in the specified speaker pattern register memory so as to recognize the input voice. |