Project AURIONAI Use-Cases Research & Innovation

FLEURS

Multilingual ASR error rate (FLEURS-102; Indonesian subset)

speechstatic
4scores
4models
0independent
0maintainer-run
4vendor-reported

Definition

Version
not in the notes
Category
speech
Maintainer
not in the notes
Status
static
Home page
not in the notes

Sources

  1. arxiv.org/pdf/2511.09690 https://arxiv.org/pdf/2511.09690
  2. Qwen3-ASR Technical Report arxiv.org https://arxiv.org/html/2601.21337v2

Scores at a Glance

Score in WER, Qwen3-ASR technical report

Lower is better for this unit.

02468
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.

Scores

4 of 4 scores shown fleurs-scores.csv
ModelScoreWho ran itSettingView or populationDateSource
Alibaba (Qwen)Qwen3-ASR-0.6B7.92 WERvendor-reportedFLEURS Indonesian, Table A.2Qwen3-ASR technical reportJan 2026arxiv.org
MetaOmnilingual ASR6.2 CERvendor-reported7B LLM-ASR, FLEURS-102, approximate–Nov 2025arxiv.org
Alibaba (Qwen)Qwen3-ASR-1.7B5.16 WERvendor-reportedFLEURS Indonesian, Table A.2Qwen3-ASR technical reportJan 2026arxiv.org
Alibaba (Qwen)Qwen3-ASR-Flash-12083.65 WERvendor-reportedFLEURS Indonesian, Table A.2Qwen3-ASR technical reportJan 2026arxiv.org
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "FLEURS", https://aurion.semai.id/benchmarks/fleurs/, records of October 2026.