The SAFE-T Corpus: A New Resource for Simulated Public Safety Communications

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6450–6457 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6450 The SAFE-T Corpus: A New Resource for Simulated Public Safety Communications Dana Delgado, Kevin Walker, Stephanie Strassel, Karen Jones, Christopher Caruso, David Graff Linguistic Data Consortium, University of Pennsylvania 3600 Market Street Suite 810, Philadelphia, PA, 19104, USA {foredana, walkerk, strassel, karj, carusocr, graff}@ldc.upenn.edu Abstract We introduce a new resource, the SAFE-T (Speech Analysis for Emergency Response Technology) Corpus, designed to simulate first- responder communications by inducing high vocal effort and urgent speech with situational background noise in a game-based collection protocol. Linguistic Data Consortium developed the SAFE-T Corpus to support the NIST (National Institute of Standards and Technology) OpenSAT (Speech Analytic Technologies) evaluation series, whose goal is to advance speech analytic technologies including automatic speech recognition, speech activity detection and keyword search in multiple domains including simulated public safety communications data. The corpus comprises over 300 hours of audio from 115 unique speakers engaged in a collaborative problem-solving activity representative of public safety communications in terms of speech content, noise types and noise levels. Portions of the corpus have been used in the OpenSAT 2019 evaluation and the full corpus will be published in the LDC catalog. We describe the design and implementation of the SAFE-T Corpus collection, discuss the approach of capturing spontaneous speech from study participants through game-based speech collection, and report on the collection results including several challenges associated with the collection. Keywords: collection, speech resources, game protocol, speech analytic technologies, transcription 1. Introduction We introduce a new resource, the SAFE-T (Speech Analysis for Emergency Response Technology) Corpus, designed to address the need for training, development and test data representing public safety communications. The SAFE-T Corpus was developed by Linguistic Data Consortium to support the National Institute of Standards and Technology OpenSAT (Speech Analytic Technologies) Evaluation campaign, and was first used as part of the OpenSAT 2019 evaluation. The goal of OpenSAT is to advance speech analytics including speech activity detection (SAD), keyword spotting (KWS) and automatic speech recognition (ASR), across multiple data domains (NIST, 2019). One especially challenging domain for speech analytic technologies is public safety communications, which are characterized by prominent background noise, radio channel noise1, speech under stress and urgent speech, the Lombard effect, and other properties. To address the need for test data in the public safety domain, LDC designed the SAFE-T (Speech Analysis for Emergency Response Technology) Corpus to collect data from speakers engaged in collaborative problem-solving activities that would be representative of public safety communications in terms of speech content, noise types and noise levels. To support OpenSAT goals, the corpus design required collection from 100 speakers recruited from the North American English speaking population, with each speaker providing a minimum of 2 hours of speech recordings, and a total audio data volume of 291 hours. The collection sought to elicit speech exhibiting specific features found in public safety communications. These include …

Content truncated to 500 words