Abstract
The filled pauses in spontaneous speech can not only be used as the basis to evaluate the fluency of speech and predict the mental state of speaker, but also affect the accuracy of automatic speech recognition system (ASR) in speech recognition.In the field of speech recognition, the detection of hesitant words in spontaneous speech is a very significant problem.The paper presents a method that an improved support vector machine (SVM) classifier and the spectral, formant and intensity of speech were used to distinguish between filled pauses and normal words in spontaneous speech of mandarin, the classification accuracy is 91.76% on the closed data set. The filled pauses and normal words in the spontaneous speech data set have been marked manually.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
