About google/fleurs/af_za
#38
by lypspeech - opened
I follow the code from evaluation example, but use the dataset named google/fleurs_af_za_test. The WER is 95, Much larger than 36.7 in paper. Why?
Are you sure the transcription wasn't stuck in a loop? Such as: https://github.com/openai/whisper/discussions/657
That could explain the bad WER
Could you share your code please @lypspeech ! Should be able to help with the set-up here!