训练数据量级

[Mddct]

请问，tokenizer的训练用了多少训练数据， 以及和whisper 对齐的数据用了多少呢？

[yongjie-lv]

每一阶段的数据都是一样的，总共38wh，中英比接近1:1

We use the same dataset for every stage, which amounts to 380k hours in total, with a roughly 1-to-1 Chinese-English split.