Novel Approaches to Speaker Clustering for Speaker Diarization in Audio Broadcast News Data
Janez ibert, France Miheli · InTech eBooks · 2008
Speech Recognition, Technologies and Applications 342the merging criteria a cross log-likelihood ratio is used.Section 3 is devoted to the development of a novel fusion-based speaker-clustering system, where the speaker segments are modeled by acoustic and prosody representations.By adding prosodic information to the basic acoustic features we have extended the standard clustering procedure in such a way that it will work with a combination of both representations.All the presented clustering procedures were assessed on two different BN audio databases and the evaluation results are presented in Section 4. Finally, a discussion of the results and the conclusions are given in Sections 5 and 6. How to referenceIn order to correctly reference this scholarly work, feel free to copy and paste the following: