fix finetuning bug

fix: padding value process bug in tokenizer decode

fix: padding bugs in 12Hz tokenizer decode.

fix: adjust finetune script for stable training.

update: update readme

update: update readme

Merge pull request #15 from vasqu/fix-fa-flags

fix loading logic

fixup wrong flags

update: modify default non_streaming_mode

fix and update: fix finetuning/prepare_data.py and update citation

fix: hf speech_tokenizer download bug.

Initial commit