The 3.54 WER should not be read as 3.54 everywhere, since averaging can hide large differences between datasets and languages. And while the model card lists Japanese, Arabic, Korean, and Russian among its supported languages, this particular leaderboard result covers only the 5
The 3.54 WER should not be read as 3.54 everywhere, since averaging can hide large difference...
The 3.54 WER should not be read as 3.54 everywhere, since averaging can hide large differences between datasets and languages. And while the model card lists Japanese, Arabic, Korean, and Russian among its supported languages, this particular leaderboard resul