Benchmark
AA-WER
A standardized speech recognition evaluation metric introduced by Artificial Analysis that unifies the audio test set and text normalization pipeline and covers multiple languages, making WER scores across different models horizontally comparable.
Timeline (last 90 days)
Oct 21
Voxtral Small Transcribe 3 has an AA-WER of 2.1%.
Unverified50%
Oct 21
AA-WER is a standardized evaluation methodology introduced by Artificial Analysis, unifying the audio test set and text normalization process, and covering multiple languages and diverse acoustic conditions.
Unverified50%