Transfer learning or Fine tuning with utf-8

I have a bilingual dataset, which is <20 hours. So, I will have to use the pre-trained Deepspeech model.

I want to use utf-8 bytes setting to either do transfer learning or fine tuning using my dataset. But I don’t have a pre-trained model that uses utf-8 bytes. Language doesn’t matter because I am using utf-8 bytes.

Can I drop the last layer in the pre-trained deepspeech model, and use 256 labels corresponding to the utf-8 bytes to further train on my dataset?

Is there any other way I can do both a) use the pre-trained deepspeech model, and b) be able to use utf-8 bytes.