Bert-VITS2-Faster

0. original algorithm repositoryBert-VITS2(v2.0.2)
1. all to trtfp16 to get 9x faster.

Using

0. after you get your own trained model. (train_ms.py)
   0.1 prepare your own 'config.yml' as the 'default_config.yml'
1. sh bert_to_onnx.sh
2. python infer_torch_export_onnx.py (first comment line: g_model_name = None)
3. see inputs.py to get trt.engine
4. python infer_backend_mix.py (test trt infer)

You can do

1. change the training forward to merge the all 6 submodel to one model to get VITS2.engine.
2. improve the performance of VITS2 on emotion.
3. when some submodel severe performance loss, you can replece the trt to onnx.