This paper discusses speaking-style and speaker adaptation techniques in a new ATR spontaneous speech database as a step toward more useful spoken dialogue translation. The task of this data concerns conversations in travel situations consisting of many forms of communications between a receptionist and a guest. Speaker-independent recognition using the data shows that the recognition performance of spontaneous speech is not very low in comparison with that of read speech when adequate speaker- and speaking-style adaptation is applied to a speaker-independent phoneme model. Furthermore, a speech recognition strategy using a context-free grammar and this speaking-style adapted acoustic model shows the effectiveness of this grammar even for spontaneous speech.
Bibliographic reference. Matsunaga, Shoichi / Kosaka, Tetsuo / Shimizu, Tohru (1995): "Speaking-style and speaker adaptation for the recognition of spontaneous dialogue speech", In EUROSPEECH-1995, 1135-1138.