DeepSData
Search site中文Contact
Dataset guide · Machine learning & corpora

2021 NIST Speaker Recognition Evaluation Development and Test Set

Developed by LDC and NIST, the 2021 NIST Speaker Recognition Evaluation Development and Test Set contains approximately 447 hours of Cantonese, Mandarin, and English conversational telephone speech, audio from video, and image data, along with answer keys, enrollment, and trial files.

← Back to dataset library · 中文版

Machine learning & corporaAccess: check the source page

Key facts

InstitutionLinguistic Data Consortium (LDC) and NIST (National Institute of Standards and Technology)
CoverageCantonese, Mandarin, and English conversational telephone speech (CTS), audio from video (AfV), and image data
Time spanPublished 2026-07-13
ScaleApproximately 447 hours of audio/video, 2 ZIP files of 41.6 GB each
LicenseTerms: check the source page
AccessUC Berkeley Library Dataverse (access by request)

Contents & fields

The dataset includes development and test speech, audio, and image data, along with answer keys, enrollment, and trial files. Data is drawn from the WeCanTalk corpus collected by LDC, containing approximately 355 hours of CTS audio (8kHz A-law sphere format), 53 hours of AfV segments (16kHz FLAC-compressed MS-WAV), 39 hours of video clips (mp4), and 202 selfie images (JPG). No field-level description is publicly available on the source page.

Research uses

Suitable for text-independent speaker recognition technology research, evaluation system development, and performance benchmarking.

Information comes from the source page. Please check that page for current details and terms.

Keywords

speaker recognitionNIST SREconversational telephone speechmultilingualevaluation dataLDC

Access & license

License: Terms: check the source page | Access: check the source page

Official site / download →

Why this is hard to get on your own

Access is restricted and requires request; license terms are custom.

Related datasets

Same domain

Need this data retrieved and prepared?

Tell us your hard requirements. We first assess availability, then retrieve for real — and if it truly cannot be obtained, we say so plainly.

Talk to usBrowse more datasets