Preprint
FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech (full text)
Alexis Conneau et al., arXiv (IEEE Spoken Language Technology Workshop 2022), 2022
Read the originalRead on October 17, 2026
- What it is
- The paper that introduced FLEURS, a speech benchmark of read sentences in 102 languages, including Japanese, Korean and Mandarin Chinese. Its text notes explain how the benchmark handles spaces between words in Chinese, Japanese and Korean, and it reports character error rate for those languages.
- How much weight it carries
- Peer-reviewed workshop paper from Google research, widely used as a benchmark. The read speech is clean and short, so scores are a floor for spontaneous speech. This PDF is the full text of the paper whose abstract page is already in the catalog.
Facts taken from this source
- The FLEURS paper says that Japanese and Korean transcripts may or may not contain spaces, depending on the transcriber, so the benchmark needed its own handling of spacing.
“Depending on the transcribers, Japanese and Ko- rean may or may not contain space irregularly.” (Section 2, transcription and tokenization)