Unverified50% confidenceBenchmarkExact time
In the KeyASRBench evaluation, the language-specific MER/WER range is 5.31% to 12.91%, while the KER range for the same set of systems is 13.61% to 33.81%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/10/2026
First Seen
Valid until: 2/8/2027
Sources
Related Entities
Related Claims
UnverifiedIn code-switching scenarios, the word error rate (WER) of current best-in-class speech recognition models remains between 15–25%, compared to within 5% for English monolingual scenarios.70% similarUnverifiedIn a balanced ASR test covering 13 dialects, the Tunisian dialect achieved a word error rate (WER) of 0.478, the highest among all dialects63% similarUnverifiedMeta Llama 2的2万亿token训练数据中英语占比约89.7%,欧洲语言合计约8%,其余所有语言共享约2%62% similarUnverifiedWithout fine-tuning, EmbeddingGemma-300M achieves only 51.7% to 60.8% recall@1 on the Belebele dataset for Tajik, Kazakh, Uzbek and Kyrgyz.61% similarUnverifiedIn the same fine-tuning, FLEURS English WER changed only from 11.04% to 10.42%, and FLEURS Arabic from 12.67% to 11.41%, so the original language capabilities degraded hardly at all.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1153739API
curl https://kongchang.com/api/v1/knowledge/claims/1153739MCP
get_claim(id=1153739)