What's the difference between "Common" and "fleur"? I don't undertand this image comparison of Large V2 vs V3?

[AIhasArrived]

[jongwook]

Hi, those are two datasets that we used for evaluating and comparing the model performance (and we didn't include the corresponding training splits in the training):

• Common Voice: https://commonvoice.mozilla.org/en/datasets
• Fleurs: https://huggingface.co/datasets/google/fleurs, https://arxiv.org/abs/2205.12446

[AIhasArrived]

Noob question, so for cantonese, large v3 is significantly better because he has a smaller WER?

[ryanheise]

Yes: Basically, smaller WER (Word Error Rate) means a smaller percentage of errors, and that's a reflection of it being better.