| 摘要 |
PURPOSE:To prevent misrecognition and increase the recognition rate by collat ing a result, outputted by a primary collation part as a series of similarity and the length of word sections, by a secondary collation part. CONSTITUTION:This method is roughly divided into the primary collation part 1 and secondary collation part 2. The primary collation part 1 analyzes and collates an input speech signal with standard patterns of all objects of recognition which are previously generated, segments input into word speeches, and outputs the result as the series of the similarity and the length of the segmented speech sections. The secondary collation part 2 utilizes the fact that the similarity to the standard pattern differ in tendency in recognized word units, inputs and handles the similarity series (consisting of similarity obtained by segmenting the input into the respective recognized object words) as the result of the primary collation as feature parameters, and recognizes a speech by pattern matching, thereby correcting an error in the recognition of the primary collation part 1. |
