Fact · Languages and accents
36percent of low-resource languages
In Omnilingual ASR, only 36 percent of the languages with less than 10 hours of training audio meet the character error rate threshold of 10 percent.
Checked against its source on October 17, 2026
In the source's own words
“On the low-resource bucket, where we have less than 10 hours of training data per language, the percentages of languages that meet the CER quality threshold fall to 34% and 36%”
Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages, Omnilingual ASR team, arXiv, 2025, Section 5.3.1.
Open the original- What it measures
- The paper groups languages by training hours. Languages with more than 50 hours form the high-resource group, and those with 10 to 50 hours form the mid-resource group.
- What it does not show
- The 34% and 36% figures come from two model variants. The authors' own test sets, not independent benchmarks.