FLEURS
A 102-language n-way parallel speech benchmark dataset by Google, with ~12 hours per language, for evaluating universal speech representations
Duration
1200 hours
Languages
102
Sample Rate
16 kHz
Published
2022-05
Description
1Speech version built upon the machine translation benchmark FLoRes-101
2N-way parallel speech data in 102 languages, covering 16 language families
3Approximately 12 hours of supervised speech data per language
4Supports multiple tasks including ASR, language identification, translation, and retrieval
5Aims to advance speech technology for low-resource languages
Language Details
| Language | Duration |
|---|---|
| Multilingual (102 languages) | 1200 hours |
Publisher