RVC + Audio separator project

Features

All you need is a model checkpoint and a song file to use as input, and the VOCAL and MR separations and even the MIX are automated.

git clone https://github.com/ksl103177/RVC_mixed_audio-separator.git

cd RVC_mixed_audio-separator

conda create -n rvc python==3.10

activate rvc

pip install -r requirements.txt

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

parameter name	Value	options
exp_dir1	Name of the model to train	None
sr2	Adjust the sampling rate value	32k, 40k, 48k
if_f0_3	Adjust whether to extract F0	True, False
spk_id5	specify speaker ID	0, 1, 2 (integer value)
save_epoch10	Specify the epoch to save the model checkpoint	10, 20, 30 (integer value)
total_epoch11	Specify the total number of training epochs	100, 200, 300 (integer value)
batch_size12	Specify the training batch size	16, 32, 48 (integer value)
in_save_latest13	Specify whether to save the latest model checkpoint	True, False
pretrained_G14	Specifies the path to the pre-trained Generato model	None
pretrained_D14	Specifies the path to the pre-trained Discriminator model	None
gpus16	Specify the GPU number used for training	0, 0,1
if_cache_gpu17	Specify whether to cache data to GPU	True, False
if_save_every_weights18	Whether to save all checkpoints	True, False
version19	Specify the model version	v1, v2
n_p	Specify the number of processes to use for model training	4, 6, 8 (integer values)
f0method	Specify F0 extraction method	pm, harvest, dio, rmvpe, rmvpe_gpu
gpus_rmvpe	Specify GPU number to use for RMVPE model	0, 0,1 (integer value)
trainset_dir4	Specify the path to the training data	None

parameter name	Value	options
trained_model_name	specifies the name of the trained model in the assets/weights path	None
file_index	Specifies the index file in the folder with the name of the trained model in the logs path	None
input_audio	Specifies the path to the song file for voice conversion	None
first_mr_output_dir	Specify the path where the MR of the song file to be voice converted is saved	None
output_info	Specify the path to the txt file where the log of the inference is recorded	None
output_audio	Specify the path where output files are saved	None
spk_id	specify the speaker ID specified in training	None
transform	Pitch transformation	-12 ~ 12 (integer value)
f0_file	Fixed to null value	None
f0_method	F0 extraction method	pm, harvest, crepe, rmvpe
index_rate	Index feature ratio	0.0 ~ 1.0 (float value)
filter_radius	Center filtering radius	1 ~ 5 (integer value)
resample_sr	Resampling sample rate	16000, 22050, 44100, 48000
rms_mix_rate	RMS mix rate	0.0 ~ 1.0 (float value)
protect	Consonant and breath protection	0.0 ~ 1.0 (float value)