LATE Dev&Test Set V1 for Latvian ASR

  Authors

Darģis, Roberts ; Znotiņš, Artūrs ; Auziņa, Ilze and Rābante-Buša, Guna

  Item identifier

http://hdl.handle.net/20.500.12574/99

 Project URL

http://www.digitalhumanities.lv/projects/vpp-late/

 Demo URL

https://late.ailab.lv

 Referenced by

https://korpuss.lv/id/LATE-mediji
https://korpuss.lv/id/LATE-sarunas

 Date issued

2024-03

 Type

audio, corpus

 Size

10 hours

 Language(s)

Latvian

 Description

A Latvian speech corpus for the development (validation), testing and comparison of ASR models. The audio data is segmented and aligned with the corresponding orthographic transcriptions which are human verified. The LATE-media subset contains both verbatim (raw) and formatted transcriptions (with punctuation, capitalisation, numbers, abbreviations, etc.), while the LATE-conversations subset currently contains only verbatim transcriptions (no punctuation, capitalisation, etc.). The dataset consists of: - 5 hours of broadcast media recordings, both spontaneous and prepared speech (2.5h dev set, 2.5h test set); - 5 hours of conversational speech recordings, spontaneous speech (2.5h dev set, 2.5h test set).

 Publisher

AiLab IMCS UL

 Acknowledgement

Ministry of Education and Science

Project code: VPP-LETONIKA-2021/1-0006

Project name: Research on Modern Latvian Language and Development of Language Technology

 Subject(s)

ASR

 Collection(s)

Language resources and tools of AiLab IMCS UL

Show full item record