Using Acoustic Diarization for Duplicate Detection

TitleUsing Acoustic Diarization for Duplicate Detection
Publication TypeTechnical Report
Year of Publication2012
AuthorsKnox, M. Tai, Friedland G., & R. Smith P.
Other Numbers3273

The following article describes the use of an acoustic diarization engine for duplicatedetection on broadcast news. Diarization is typically used to partition audio into speakerhomogeneous regions, or in other words, to determine “who spoke when.” In thissetting, however, we use diarization to segment the recordings and group the segmentsinto homogeneous clusters. Diarization is performed both on the full length broadcastnews recordings as well as the short clips (which we are classifying as either a duplicateor not). We then compare the similarity of models trained on the clusters to determinewhether the time allocated to the cluster from the short clip is from the originalbroadcast news recording, or a duplicate. We tested our system under a variety of audioconditions: unmodified, with reverberation, resampled, and lowpass filtered. On ourtest set, the areas under the receiver operating characteristic curve for the audioconditions were 0.91, 0.89, 0.61, and 0.64 respectively.

Bibliographic Notes

ICSI Technical Report TR-12-005

Abbreviated Authors

M. Knox, G. Friedland, and R. P. Smith

ICSI Research Group


ICSI Publication Type

Technical Report