next up previous
Next: 6. Conclusions Up: 5. Synthesis Experiment Previous: Synthesis experiment

Evaluation

One hundred city or town names were selected from among the 3,167 synthesized names so as to be unfamiliar to the ten subjects participating in the evaluation. The same 100 names were recorded by the original narrator. First, the 100 synthesized words, recorded on DAT tape and played back via headphones, were presented to the subjects who attempted to write the names down as a test of clearness. The same test was then repeated using the natural words in different order. An opinion test was conducted by presenting each subject with the natural and synthesized words (total of 200) in random order. Each word was assigned a score ranging from 1(worst) to 5(best). As is shown in Table 2,
   
Table 2: The result of evaluation
    Average of
  Clearness opinion scores
Natural words 90.4 % 4.6
Synthesized words 91.5 % 3.5

the synthesized words were comparable in quality to the natural words. Considering that these synthesized words will be used to confirm the city or town names input in an interactive process like a directory assistance service, this level of quality is good enough to ensure customer satisfaction.
next up previous
Next: 6. Conclusions Up: 5. Synthesis Experiment Previous: Synthesis experiment
Jin'ichi Murakami
2000-01-17