| 摘要 |
PURPOSE:To improve recognition performance by obtaining the results of recognition by making overall decision on the results of the recognition based on the approximate characteristics by collation of continuous patterns and the results of recognition based on the detail characteristics by collation of patterns after the detection of beginning and terminal ends. CONSTITUTION:A 1st analyzing section 4 extracts the 1st characteristic parameters of the low number of dimensions used for the collation processing of the continuous patterns by the non-fixing of the beginning and terminal ends from input voices and a 2nd analyzing section 5 extracts the 2nd voice characteristic parameter of the high number of dimensions used for the collation processing patterns utilizing the detail characteristics of the voice sections. While a recognition processing section 3 executes the collation of the patterns with a recognition dictionary 6 continuously while the beginning and terminal ends are held non- fixed, the beginning and terminal ends of the input voice are detected. This section determines the voice characteristic vectors between the beginning and terminal ends and executes the collation of the patterns with the recognition dictionary 6. The final results of the recognition to the input voices are obtd. by the overall processing of the respectively obtd. results of the recognition in a recognition result processing section 11. The recognition performance and more particularly the reject performance to unnecessary words are sufficiently enhanced in this way. |
