Paper: Look Who is Talking: Soundbite Speaker Name Recognition in Broadcast News Speech

ACL ID N07-2026
Title Look Who is Talking: Soundbite Speaker Name Recognition in Broadcast News Speech
Venue Human Language Technologies
Session Short Paper
Year 2007
Authors

Speaker name recognition plays an important role in many spoken language applications, such as rich transcription, information extrac- tion, question answering, and opinion mining. In this paper, we developed an SVM-based classification framework to determine the speaker names for those included speech seg- ments in broadcast news speech, called sound- bites. We evaluated a variety of features with different feature selection strategies. Experi- ments on Mandarin broadcast news speech show that using our proposed approach, the soundbite speaker name recognition (SSNR) accuracy is 68.9% on our blind test set, an ab- solute 10% improvement compared to a base- line system, which chooses the person name closest to the soundbite.