Selection of recognition candidates based on parallel recognition of sentences and partial words in spoken dialogue
Makiho Kawakami, Hiromasa Terashi, Masafumi Nishida, Yasuo Horiuchi, Akira Ichikawa · The Journal of the Acoustical Society of America · 2006
We have proposed a method that predicts a user’s utterances in spoken dialogue systems by recognizing prediction sentences of each dialogue state, thereby decreasing recognition errors. However, the conventional method might repeat recognition errors because it confirms the whole sentence even if it recognizes only a part of a long word, such as a compound word, correctly. For this study, we introduce a method using decoders based on prediction sentences and partial words obtained by dividing long words. The proposed method can confirm a user’s utterances by selecting a candidate to a partial matched word using recognition results of the prediction sentence and a divided partial word. We conducted experiments for 560 utterances of seven persons using a navigation system. Results showed that the word recognition accuracy was 81.1%. The rate at which recognition results by the prediction sentence and a divided partial word were an exact or partial match was 70.5%. The rate of recognition errors when the recognition results were an exact or partial match was 0.9%. Therefore, we demonstrated that it is possible to confirm a user’s utterances using the proposed method by selecting recognition candidates.