Hi,
I had trained 10 hours own data which consist of certain keywords and did a transfer learning on DeepSpeech 0.8.2 checkpoint for around 25 epoch with early stopping and while comparing the results of generated model (with own data) and pre-trained model of DeepSpeech 0.8.2 on test dataset own model has a hude drop in general accuracy while it could identify certain keywords.
Question is there a way i could train the keywords data on the checkpoint without losing its general accuracy.