• Publications
  • Influence
TOWARDS ROBUST SPEAKER SEGMENTATION: THE ICSI-SRI FALL 2004 DIARIZATION SYSTEM
TLDR
The ICSI-SRI system is an agglomerative clustering system that uses a BIC-like measure to determine when to stop merging clusters and to decide which pairs of clusters to merge, providing robustness and portability.
An Analysis of Sentence Segmentation Features for Broadcast News , Broadcast Conversations , and Meetings
TLDR
This work analyzes sentence segmentation performance as a function of feature types and transcription (manual versus automatic) for news speech, meetings, and a new corpus of broadcast conversations to show that features for broadcast news transfer well to meetings and broadcast conversations.
The ICSI+ multilingual sentence segmentation system
TLDR
It is demonstrated that the proposed methodology including pitch- and energyrelated prosodic features performs significantly better than a baseline system that uses words and simple pause features only.
Cross-linguistic analysis of prosodic features for sentence segmentation
TLDR
Experiments in Arabic, English, and Mandarin show that prosodic features make significant contributions in all three languages, and feature selection results indicate that feature relevancy can vary greatly depending on the target language, and therefore the optimal feature subset varies considerably between languages.
Prosodic Similarities of Dialog Act Boundaries Across Speaking Styles
Spontaneous speech differs in a multitude of ways from read, formal, or laboratory speech (Maclay & Osgood 1959, Goldman-Eisler 1968, Levelt 1983, Biber 1988, Howell & Kadi-Hanifi 1991, Eskenazi
Entropy Based Classifier Combination for Sentence Segmentation
TLDR
A new dynamic entropy-based classifier combination approach is proposed to combine these classifiers, and it is compared with the traditional classifiers combination techniques, namely, voting, linear regression and logistic regression.
Cross-Genre Feature Comparisons for Spoken Sentence Segmentation
TLDR
Comparing sentence segmentation for speech from broadcast news versus natural multi-party meetings, using identical lexical and prosodic feature sets across genres shows that features sets can be reduced with little or no loss in performance, and the contribution of different feature types differs significantly by genre.
Cross-Genre Feature Comparisons for Spoken Sentence Segmentation
TLDR
Comparing sentence segmentation for speech from broadcast news versus natural multi-party meetings, using identical lexical and prosodic feature sets across genres shows that features sets can be reduced with little or no loss in performance, and the contribution of different feature types differs significantly by genre.
Cross-Genre Feature Comparisons for Spoken Sentence Segmentation
TLDR
Comparing sentence segmentation for speech from broadcast news versus natural multi-party meetings, using identical lexical and prosodic feature sets across genres shows that features sets can be reduced with little or no loss in performance and the contribution of different feature types differs significantly by genre.
Automatic Design of Prosodic Features for Sentence Segmentation
TLDR
This study finds that the HLDA system can perform about as well as the baseline features, which suggests that an automatic approach to learning good features for a new language may be of assistance.
...
1
2
...