A Novel Detection Method of Filled Pause in Mandarin Spontaneous Speech

Filled pause is one of the hesitation phenomena that current speech recognizers can not effectively handle. Detecting filled pauses is important in spontaneous speech dialogue systems because they play valuable roles, such as helping a speaker keep a conversational turn, in oral communication. In this paper, a novel detection method for filled pause is proposed based on a strategy of two-level decisions. Firstly, hypothetical filled pauses are extracted from non-silence signal segments. Then HMMs are trained and employed to recognize filled pauses from hypothetical filled pauses. Experiment results show that the average precision and recall rate of filled pauses are 80.66% and 92.59% respectively. Moreover, this filled pause detector can distinguish filled pauses from elongated words, which are not achieved in the previous works.