How is wer?
#1
by Shekharmeena - opened
how is the accuracy of the model
It needs more training and I also probably need to work on what works best for inference as well. It's inconsistency makes me think it's not worth benchmarking at the moment because it's very accurate for some sentences but for others it 100% inaccurate for others. I've noticed this inconsistency seems to be related to the length of the target audio but more experimentation is needed.
I still plan on running a few benchmarks on it since someone's mentioned it now, so look out for that