Speech and Vision Lab

In this paper we demonstrate an automatic summarization system for broadcast news shows. The proposed technique does not require ASR transcripts or human reference summaries. The system exploits the role of anchor speaker in a news show by tracking his/her speech to construct indicative extractive summaries. Speaker tracking is done by autoassociative neural network model. Summaries are generated for desired compression ratio . The output summary is presented in the form of speech. The experiments were carried out on BBC news podcasts available online. The evaluation results show that summarization of structured speech documents like broadcast news shows can be performed with good accuracies comparable to text summarization.