Call My Net 2: A New Resource for Speaker Recognition

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6621–6626 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6621 Call My Net 2: A New Resource for Speaker Recognition Karen Jones, Stephanie Strassel, Kevin Walker, Jonathan Wright Linguistic Data Consortium University of Pennsylvania, Philadelphia, USA {karj, strassel, walkerk, jdwright}@ldc.upenn.edu Abstract We introduce the Call My Net 2 (CMN2) Corpus, a new resource for speaker recognition featuring Tunisian Arabic conversations between friends and family, incorporating both traditional telephony and VoIP data. The corpus contains data from over 400 Tunisian Arabic speakers collected via a custom-built platform deployed in Tunis, with each speaker making 10 or more calls each lasting up to 10 minutes. Calls include speech in various realistic and natural acoustic settings, both noisy and non-noisy. Speakers used a variety of handsets, including landline and mobile devices, and made VoIP calls from tablets or computers. All calls were subject to a series of manual and automatic quality checks, including speech duration, audio quality, language identity and speaker identity. The CMN2 corpus has been used in two NIST Speaker Recognition Evaluations (SRE18 and SRE19), and the SRE test sets as well as the full CMN2 corpus will be published in the Linguistic Data Consortium Catalog. We describe CMN2 corpus requirements, the telephone collection platform, and procedures for call collection. We review properties of the CMN2 dataset and discuss features of the corpus that distinguish it from prior SRE collection efforts, including some of the technical challenges encountered with collecting VoIP data. Keywords: speaker recognition, speech database, telephony, VoIP, data centers, Tunisian Arabic 1. Introduction The Call My Net 2 (CMN2) Corpus is a large collection of Tunisian Arabic conversational telephone speech (CTS) recordings that was created to support the advancement of technology in the field of text independent speaker recognition. CMN2 adds a new language to LDC’s SRE collections since Tunisian Arabic has not specifically been collected in the past. Also, in addition to traditional telephony data the CMN2 corpus contains VoIP data for the first time. All call recordings were collected in Tunis and the creation of this corpus involved developing several new infrastructural solutions to hosting and managing a remote collection. In total, 4,562 calls were made by 408 Tunisian Arabic speakers who made at least 10 calls each from December 2016 to January 2018. 2306 recordings from 213 speakers were used in the NIST 2018 Speaker Recognition Evaluation (NIST, 2018; Sadjadi et al., 2019). Calls from the remaining speakers were reserved for the NIST 2019 Speaker Recognition Evaluation (NIST, 2019). 2. Language Together, the varieties of Arabic spoken throughout the Maghreb comprise a dialect continuum that encompasses the everyday Arabic dialects spoken in Libya, Tunisia, Algeria, Morocco, Western Sahara and Mauritania. While Maghrebi Arabic has featured in prior LDC collections such as LRE11, the regional dialects were not specifically distinguished. In contrast, one of the requirements for the CMN2 collection was that all study participants be native or highly …

Content truncated to 500 words