Bert-VITS2-Faster

Using

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

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

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
bert		bert
configs		configs
emotional/wav2vec2-large-robust-12-ft-emotion-msp-dim		emotional/wav2vec2-large-robust-12-ft-emotion-msp-dim
filelists		filelists
monotonic_align		monotonic_align
text		text
tools		tools
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
attentions.py		attentions.py
bert_gen.py		bert_gen.py
bert_to_onnx.sh		bert_to_onnx.sh
commons.py		commons.py
config.py		config.py
data_utils.py		data_utils.py
default_config.yml		default_config.yml
emo_gen.py		emo_gen.py
infer_.py		infer_.py
infer_backend_mix.py		infer_backend_mix.py
infer_torch_export_onnx.py		infer_torch_export_onnx.py
inputs.py		inputs.py
losses.py		losses.py
mel_processing.py		mel_processing.py
models.py		models.py
modules.py		modules.py
onnx_to_trt.sh		onnx_to_trt.sh
preprocess_text.py		preprocess_text.py
re_matching.py		re_matching.py
requirements.txt		requirements.txt
resample.py		resample.py
train_ms.py		train_ms.py
transforms.py		transforms.py
update_status.py		update_status.py
utils.py		utils.py