摘要 |
PURPOSE:To make a simple and reliable dictionary correction possible while allowing to cancel only the correcting operation caused by mistaken input patterns by providing a dictionary correction processing means and a restoration information storage means. CONSTITUTION:The voice pattern introduced through a voice parameter converting part 11 is sent to a voice recognition part 13 and a dictionary correction processing part 15 where the number of templates with respect to the input voice pattern is checked. When the ;preset storage number is two, the input voice parameter is collated to the parameters of two templates read from a dictionary 12 to find a template analogous to the input voice pattern. The mean value between the parameters of the input voice pattern and the template is found to give a covected template parameter while the precorrection parameters and the management information are evacuated in a restored information storage part 17. The template which was the object of correction is deleted from the dictionary. By this, the dictionary correction can be performed simply and at a higher speed.
|