Its purpose is to train Tokushuhaku, a complex phenomenon which is still incompletely understood. Loosely speaking, training based on model speech is more suitable for Tokushuhaku development. The tolerance range of each mora rhythm in a given word is used to check whether a learner uttered rhythmically or not. Conventional feedback methods offer the feedbacks as like ``say it longer'' or ``say it shorter''. Instead, we offer a visualized feedback method that every learner has become accustomed to. Figure 5.17 shows a typical example of the user interface, in this case a non-native learner A uttered the target word of [oqto] (husband).
The three main windows are ( from top to bottom ) the model window, learner window and control panel. The model window displayed the target speech rhythm to be achieved by the learners, along with the waveform and segmented information. On the other hand, the learner window displayed the results of analyzing learner A's speech. In this example, he has pronounced the mora obstruent /q/ very shortly so that it sounds like the word [oto] which means sound. Consequently, the wrongly pronounced mora are highlighted with a shadow box. By clicking the PLAY button on the control panel, he can listen to his attempt's speech or model's exact speech.
We had let him practice it using our system for a while, and then verified how his rhythmical error has been reduced. Figure 5.18 provides an illustrative example of his improved rhythm of the word [oqto] after self-feedback. It is found out that his mora obstruent /q/ has been longer and his rhythm is located in the normal range. The improvement of the rhythm error was also verified by human judgement. The lesson presented here is just one example of how the system can be used as an application of CAPL.
図 5.17: Example of user interface for training based on model speech.
図 5.18: Improved speech rhythm of learner A for the [oqto] after self-feedback
