Personal tools
You are here: Home Knowledge Base Unsupervised Speaker Clustering Using a Global Similarity and F0 Features  

Unsupervised Speaker Clustering Using a Global Similarity and F0 Features

Main information
This paper investigates an unsupervised speaker clustering approach that exploits global similarity and also proposes extending the standard cepstal feature set used for speaker clustering with prosodic features, extracted from F0. The global-similarity-based speaker clustering algorithm, initially proposed by the authors in [6], leverages the insight that audio segments within a single cluster are not only similar to one another, but also display the same patterns of similarities and differences with audio segments belonging to all other clusters. First, speaker clustering performance using the standard Bayesian Information Criterion (BIC) is compared to the performance achieved using a BIC-based algorithm incorporating global similarity. Then both clustering techniques are tested using an extended feature set including F0-derived features in addition to the standard cepstral features. The evaluation, which is performed on data recorded from German language radio, shows the clear benefits of using global information when performing clustering. It also demonstrates that in most cases F0-features outperform the cepstral feature set both in standard BIC clustering and in the BIC global-similarity-based approach.  
WP5: Detection, Extraction and Annotation of Knowledge. 
IAIS  Konstantin Biatov, Martha Larson 
2007-03-15 16:16  Request for more detail


Access and Use Rights
Condition of use defined in response to "need to access request". 
Copyright Fraunhofer Institut Intelligente Analyse- und Informationssysteme. 
 
Closed, attachment is not public 


This item is not available for public download. For further information click on Request for more details above

LIVE's knowledge base
  • What is available here?

Here you can search and access the documented work of the LIVE project. Documents include the official reporting of the project's work known as "deliverables", scientific and technical publications, study reports, press material, presentation material  as well as IPR pending material.

  • How can I access the material?

You can search this knowledge base by project partner, work area or key words.

By simply clicking "all" the complete list of documents will be listed.

  • Is everything open to the public?

Please note that only those publications marked in green have been cleared for immediate public access. To request access to those publications not open please click on the "request for more detail" link in the relevant document. Your request will be automatically sent to the authors for consideration.