Team ELISA System for DARPA LORELEI Speech Evaluation 2016

Papadopoulos, Pavlos; Travadi, Ruchir; Vaz, Colin; Malandrakis, Nikolaos; Hermjakob, Ulf; Pourdamghani, Nima; Pust, Michael; Zhang, Boliang; Pan, Xiaoman; Lu, Di; Lin, Ying; Glembek, Ondřej; Baskar, Murali Karthick; Karafiát, Martin; Burget, Lukáš; Hasegawa-Johnson, Mark; Ji, Heng; May, Jonathan; Knight, Kevin; Narayanan, Shrikanth S.

doi:10.21437/Interspeech.2017-180

Team ELISA System for DARPA LORELEI Speech Evaluation 2016

Pavlos Papadopoulos, Ruchir Travadi, Colin Vaz, Nikolaos Malandrakis, Ulf Hermjakob, Nima Pourdamghani, Michael Pust, Boliang Zhang, Xiaoman Pan, Di Lu, Ying Lin, Ondřej Glembek, Murali Karthick Baskar, Martin Karafiát, Lukáš Burget, Mark Hasegawa-Johnson, Heng Ji, Jonathan May, Kevin Knight, Shrikanth S. Narayanan

In this paper, we describe the system designed and developed by team ELISA for DARPA’s LORELEI (Low Resource Languages for Emergent Incidents) pilot speech evaluation. The goal of the LORELEI program is to guide rapid resource deployment for humanitarian relief (e.g. for natural disasters), with a focus on “low-resource” language locations, where the cost of developing technologies for automated human language tools can be prohibitive both in monetary terms and timewise. In this phase of the program, the speech evaluation consisted of three separate tasks: detecting presence of an incident, classifying incident type, and classifying incident type along with identifying the location where it occurs. The performance metric was area under curve of precision-recall curves. Team ELISA competed against five other teams and won all the subtasks.