Zipformer MVQ

[marcoyang1998]

This PR makes training with knowledge distillation an option in the Zipformer recipe. The knowledge distillation method is MVQ-KD.

The teacher targets can be downloaded via the following command:

./distillation_with_hubert.sh --stage 2 --stop_stage 2

To turn on knowledge distillation, you will need to set --enable-distillation True. It is applicable to both streaming and non-streaming Zipformers.

Detailed results will follow.

[marcoyang1998]

Some results:

100 hours:

model	test-clean	test-other
baseline, epoch-30-avg-9	5.97	15.73
+ mvq, epoch-30-avg-9	5.13	13.08

960 hours:

model	test-clean	test-other
baseline, epoch-30-avg-9	2.25	5.06
+ mvq, epoch-30-avg-9	2.18	4.86