next up previous
次へ: Discussion ( Combining Viterbi 上へ: Continuous Speech Experiments 戻る: Experimental conditions

Results of Experiments

We show the results of our experiments in Table 2. The table shows that forward decoding is superior to Viterbi decoding for the word accuracy rate. However, it is inferior for the word correct rate. These results mean that forward decoding has fewer insertion errors than Viterbi decoding.


表 2: Experimental results
Forward decoding text-open text-closed
word correct 34.0% 36.3%
word accuracy 25.9% 29.0%
Viterbi decoding text-open text-closed
word correct 54.1% 56.8%
word accuracy 20.1% 23.6%



Jin'ichi Murakami 平成13年1月19日