図 4.3: Spectrographical comparison between before-feedback and after-feedback
The system performance needs to be evaluated for a long term period of time. Prior to that, preliminary training was completed for evaluating the effectiveness of the CAPL system. First, the learner B was taught the correct articulatory configuration for /d/ and /t/ sounds through the feedback instruction, and then we have investigated his /t/ sound again. We found that his /t/ sound was improved compared with the previous trial. To verify this, the spectrographic representations of before- and after-feedback were shown in Figure 4.3 for the target word [kyooto]. In fact, the distinctive contrast is not so clear for the /t/ sound, but the formant tracks of other vowel sounds are so much changed. It is believed that the target consonant sound affected the preceding as well as the following vowel sounds. As for the pronunciation errors described in the previous subsection, we had the learners A and B practice each target phoneme /a/ and /sh/ of the test-set. Figure 4.7 and Table 4.4 show the improved results after the feedback instruction.
図 4.3: Spectrographical comparison between before-feedback and after-feedback
図 4.4: An example of consonant feedback using articulatory configuration
図 4.5: An example of consonant feedback based on Pair-Wise classifiers
図 4.6: An example of vowel feedback by plotting 
on vowel diagram
図 4.7: plots in the vowel diagram after feedback
It is realized that the samples of phoneme /a/ are all located within
the loci of native speakers' /a/, and the majority of
the samples of phoneme /sh/ are classified into post-alveolar,
not unvoiced dental. The mean score of phoneme /a/ increased
from -27.96 to -26.19, and that of phoneme /sh/ also
improved from -27.49 to -26.82, respectively.
| /sh/ | trial | VLAB | ULAB | VDEN | UDEN | POST | PALA | VVEL | UVEL | GLOT | |
| POST | 8 | 6 | 2 |
| /sh/ | trial | PLOS | PLOY | NASA | NASY | FRIC | FRIY | AFFR | TAPF | TAPY | APPR | |
| FRIY | 8 | 1 | 7 |
(It displays the classified result for the improved /sh/ sounds that the learner B produced after feedback. The upper table is the result of the place of articulation and the lower table is that of the manner of articulation.)
