5.2.2 Our Training Approach

The suprasegmental errors that have been examined most extensively in relation to intelligibility have been involved with speech rhythm. Many of these errors appear to be due to poor speech rhythm control. Many attempts to determine the relationship between deviant speech rhythm and intelligibility have been found in recent research. Studies that have attempted to determine the cause and effect relationship between speech errors and intelligibility have dealt primarily with this speech rhythm. They found that sentences spoken with correct rhythm were substantially more intelligible than those that were not.

Figure 5.1 illustrates the block diagram for speech processing procedures required for the automatic training system for speech rhythm. As the native speakers' speech database, the phonetically balanced word database of Advanced Telecommunications Research (ATR) Institute, which contains 20 native male speakers' 520 word utterances, was used. For the Hard and Soft Template, the native speakers' rhythms were separately calculated for their means and standard deviations beforehand, and they are applied to evaluate the speech rhythm of learners. The feedback instruction is provided by representing mora duration on the screen as well as replaying a learner's speech, thus the feedback such as rhythmically longer or rhythmically shorter is given to them.


next up previous contents
Next: 5.3 Analysis of Japanese Up: 5.2 Approach to Speech Previous: Tokushuhaku Intelligibility

Jo Chul-Ho
Wed Oct 13 17:59:27 JST 1999