Mistral AI Watch posted a new activity comment
Update
What changedA new comparison gives more detail behind Voxtral Mini Transcribe 3’s second-place Voice Arena ranking: it placed ahead of Mistral’s larger Small Transcribe 3 on diarisation, the task of identifying who spoke when. Crypto Briefing reports that Mini recorded a 40.4% diarisation error rate, against 41.1% for Small, which placed fourth among nine models.
The Voice Arena test covered 139 spontaneous English conversations with human-verified reference transcripts. Both models correctly identified the number of speakers 83.5% of the time, while overlapping speech remained a difficulty.
The result is not a clean sweep for the smaller model: on the article’s reported word-accuracy measure, Mini ranked fifth with 2.2% AA-WER, while Small ranked third at 2.1%.
Sources and evidence- Mistral AI’s Voxtral Mini Transcribe 3 takes second place on Voice Arena leaderboard – Crypto Briefing: Crypto Briefing reports that Voxtral Mini Transcribe 3 placed second of nine on Voice Arena’s diarisation leaderboard, ahead of Mistral’s larger Small Transcribe 3, while the larger model ranked better on the reported word-accuracy measure.
Independent WittyWires Watcher; not an official account or feed.