This wiki presents the LIUM_SpkDiarization tools. LIUM_SpkDiarization is a software dedicated to speaker diarization (ie speaker segmentation and clustering). It is written in Java, and includes the most recent developments in the domain.

LIUM_SpkDiarization comprises a full set of tools to create a complete system for speaker diarization, going from the audio signal to speaker clustering based on the CLR/NCLR metrics. These tools include MFCC computation, speech/non-speech detection, and speaker diarization methods.

This toolkit was developed for the French ESTER2 evaluation campaign, where it obtained the best results for the task of speaker diarization of broadcast news in 2008 [1]. Please note that the toolbox is optimized for radio or tv shows. You should not expect the same level of performances on phone conversation and meetings.

If you are using this toolkit in your research please cite one of these papers.

Speaker Diarization

 M. Rouvier, G. Dupuy, P. Gay, E. Khoury, T. Merlin, S. Meignier, “An Open-source State-of-the-art Toolbox for Broadcast News Diarization,“ Interspeech, Lyon (France), 25-29 Aug. 2013

 S. Meignier, T. Merlin, “LIUM SpkDiarization: An Open Source Toolkit For Diarization,” in Proc. CMU SPUD Workshop, March 2010, Dallas (Texas, USA).

Speaker Identification

V. Jousse, S. Petitrenaud, S. Meignier, Y. Estève and C. Jacquin, “Automatic named identification of speakers using diarization and ASR systems,” in Proc. ICASSP 2009, 19-24 April 2009, Taipei (Taiwan)

[1] S. Galliano, G. Gravier, and L. Chaubard, “The ESTER 2 evaluation campaign for the rich transcription of French radio broadcasts,” in Interspeech 2009, September 2009.