Publication Details
Title: Acoustic Change Detection and Clustering on Broadcast News
Author: J. Ferreiros and D. Ellis
Group: ICSI Technical Reports
Date: March 2000
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/2000/tr-00-006.pdf
Overview:
We have developed a system that breaks input speech into speech segments using an acoustic similarity measure between two segments. The aim is to detect the time points where the acoustic characteristics change. These changes are caused mainly by speaker changes but also by acoustic environment changes. We have also developed another system that performs a clustering of the speech chunks generated by the former system and creates clusters containing the segments with homogeneous acoustic conditions. This clustering is fed back to the acoustic change detector to make more robust decisions based on both the acoustic similarity measurement between two consecutive segments and using extra information coming from the distance between the two clusters to which each of them belong. The interaction between the two systems (acoustic change detection and clustering) improves the results obtained for both aims.
Bibliographic Information:
ICSI Technical Report TR-00-006
Bibliographic Reference:
J. Ferreiros and D. Ellis. Acoustic Change Detection and Clustering on Broadcast News. ICSI Technical Report TR-00-006, March 2000
Author: J. Ferreiros and D. Ellis
Group: ICSI Technical Reports
Date: March 2000
PDF: ftp://ftp.icsi.berkeley.edu/pub/techreports/2000/tr-00-006.pdf
Overview:
We have developed a system that breaks input speech into speech segments using an acoustic similarity measure between two segments. The aim is to detect the time points where the acoustic characteristics change. These changes are caused mainly by speaker changes but also by acoustic environment changes. We have also developed another system that performs a clustering of the speech chunks generated by the former system and creates clusters containing the segments with homogeneous acoustic conditions. This clustering is fed back to the acoustic change detector to make more robust decisions based on both the acoustic similarity measurement between two consecutive segments and using extra information coming from the distance between the two clusters to which each of them belong. The interaction between the two systems (acoustic change detection and clustering) improves the results obtained for both aims.
Bibliographic Information:
ICSI Technical Report TR-00-006
Bibliographic Reference:
J. Ferreiros and D. Ellis. Acoustic Change Detection and Clustering on Broadcast News. ICSI Technical Report TR-00-006, March 2000