论文部分内容阅读
This paper presents a deep neural network(DNN)approach to sentence boundary detection in broadcast news.We extract prosodic and lexical features at each inter-word position in the transcripts and learn a sequential classifier to label these positions as either boundary or nonboundary.This work is realized by a hybrid DNN-CRF(conditional random field)architecture.The DNN accepts prosodic feature inputs and non-linearly maps them into boundary/non-boundary posterior probability outputs.