In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 47.03 percent on Common Voice Arabic.

Quote found word for word on the live source, last checked October 19, 2026

In the source's own words

“Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27”

Voxtral, Mistral AI, arXiv (Audio and Speech Processing), 2025, Table 5 (Common Voice), column ar.

What it measures: Common Voice is speech recorded by volunteers, so its audio is less controlled than read benchmarks. The paper names the version Common Voice 15.1. The table sits in the same paper, so the systems were compared on the same test.
What it does not show: One paper's runs under its own text normalization and decoding, so compare systems within this table, not with other papers' numbers. The paper gives no date per row. The paper names this system "ElevenLabs Scribe" in Table 3 and "Scribe" in Tables 4 to 6; this row treats them as one system. Arabic is left out of the paper's Common Voice macro-average because every model scores above 45 percent on it.

A row of Speech Recognition Accuracy: Word Error Rates, as ElevenLabs Scribe, CV Arabic: 47.03 percent.