Graduate School and Research Center in Digital Sciences

The EURECOM submission to the first DIHARD challenge

Patino, Jose; Delgado, Héctor; Evans, Nicholas

INTERSPEECH 2018, 19th Annual Conference of the International Speech Communication Association, September 2-6, 2018, Hyderabad, India

The first DIHARD challenge aims to promote speaker diarization research and to foster progress in domain robustness. This paper reports EURECOM’s submission to the DIHARD challenge. It is based upon a low-resource, domain-robust binary key approach to speaker modelling. New contributions include the use of an infinite impulse response - constant Q Melfrequency cepstral coefficient (ICMC) front-end, a clustering selection / stopping criterion algorithm based on spectral clustering and a mechanism to detect single-speaker trials. Experimental results obtained using the standard DIHARD database show that the contributions reported in this paper deliver relative improvements of 39% in terms of the diarization error rate over the baseline algorithm. An absolute DER of 29% on the evaluation set compares favourably with those of competing systems, especially given that the binary key system is highly efficient, running 63 times faster than real-time.

Document Bibtex

Title:The EURECOM submission to the first DIHARD challenge
Keywords:Speaker diarization, binary key, spectral clustering, zero-resource diarization
Type:Conference
Language:English
City:Hyderabad
Country:INDIA
Date:
Department:Digital Security
Eurecom ref:5581
Copyright: © ISCA. Personal use of this material is permitted. The definitive version of this paper was published in INTERSPEECH 2018, 19th Annual Conference of the International Speech Communication Association, September 2-6, 2018, Hyderabad, India and is available at :
Bibtex: @inproceedings{EURECOM+5581, year = {2018}, title = {{T}he {EURECOM} submission to the first {DIHARD} challenge}, author = {{P}atino, {J}ose and {D}elgado, {H}{\'e}ctor and {E}vans, {N}icholas}, booktitle = {{INTERSPEECH} 2018, 19th {A}nnual {C}onference of the {I}nternational {S}peech {C}ommunication {A}ssociation, {S}eptember 2-6, 2018, {H}yderabad, {I}ndia}, address = {{H}yderabad, {INDIA}}, month = {09}, url = {http://www.eurecom.fr/publication/5581} }
See also: