This study proposes a new evaluation method for utterance-listening system performances verifying the effectiveness of an application of the disparity method to acquire one of the basic methods for the word intelligibility study from an engineer's view point.
The effectiveness of the disparity method was inspected to answer the following questions; (1) How to analyze and evaluate the utterances and the listenings by the disparity method? (2) Can the errors be compared with machine recognition performances? and how to describe the differences between the machine recognition, utterances, and listenings? (3) How to describe similar word pair characteristics by the method in the case of human performances? (4) How to describe and evaluate individual performances from an engineer's view point.
Erroneous utterances were picked up from recorded data, and erroneously written down words obtained from dictation were picked up from the word lists for the disparity calculation. The solution on all items of the questions listed above were given by the disparity method. Thus the effectiveness of the disparity method for the evaluation of human utterance-listening system was confirmed through the survey. This study might only be monochromatic survey while the human behavior is colorful in their daily aspects. It must be emphasized that further interdisciplinary cooperation for finding out the systematic method not only for the word intelligibility study but the aural communication study.
View full abstract