15th International Congress of Phonetic Sciences (ICPhS-15)
Yes-No answers have traditionally posed major difficulties in speech
recognition: the amount of signal is too short and may lead to mis-recognition.
We think that the speech signal itself, that is, intonation (F0 and
energy) are not enough to fulfil this task. When the information the
signal can provide is so little, we must look for other sources. This
paper analyzes intonation and gestures to process yes-no answers in
real dialogues in a human-machine system. This study is especially
interesting when non-lexical items are employed to substitute the words
"yes" and "no" in casual spontaneous speech.
We have studied 50 dialog extracts in Spanish. In our experiments trained evaluators report on the speakers' answers. The evaluators relied mostly on gestures when the information was not clear.
Bibliographic reference. López-Soto, Teresa (2003): "Intonation vs. gestures in yes-no answers", In ICPhS-15, 1237-1240.