Skip to content

Latest commit

 

History

History
85 lines (66 loc) · 2.72 KB

usage.md

File metadata and controls

85 lines (66 loc) · 2.72 KB
Raw

aidio wrapper

Singing Voice Separation

python aidio.py features --features_path E:\aidio_data\features --feature svs_openunmix --raw_path E:\parsed_singers.v2

Frame Selection

python aidio.py features --features_path E:\aidio_data\features --feature frame_selection --raw_path E:\parsed_singers.v2

Train GMM

python aidio.py model --model gmm --experiment svs_2 --data_path /home/voyanedel/data/data/2d/svs-bin-full --label_filename labels.mfcc.csv

Train Wavenet

Wavenet Vanilla

python aidio.py model --model wavenet --data_path /home/vichoko/data/data/1d/svs-bin-full --label_filename labels.csv --gpus [0] --experiment foo

Wavenet Transformer

python aidio.py model --model wavenet --data_path /home/vichoko/data/data/1d/svs-bin-full --label_filename labels.csv --gpus [0,1] --experiment foo

Wavenet BiLSTM

python aidio.py model --model wavenet_lstm --data_path /home/vichoko/data/data/1d/svs-bin-full --label_filename labels.csv --gpus [0] --experiment foo

GMM

python aidio.py model --model gmm --data_path /home/vichoko/data/data/2d/svs-bin-full --label_filename labels.csv --gpus [0] --experiment foo

clean dataset

python commands\clean_dataset.py --src_path E:\aidio_data\features\svs_openunmix --dest_path E:\aidio_data\features\svs_ou_skipped --label_file labels.svs_openunmix.csv

optimize dataset

python commands\optimize_dataset.py --raw_path E:\parsed_singers.v2

Known Errors

If you get

Error: mkl-service + Intel(R) MKL: MKL_THREADING_LAYER=INTEL is incompatible with %s library."
                            "\n\tTry to import numpy first or set the threading layer accordingly.

Then export this variable

export MKL_THREADING_LAYER=GNU

1D to 2D

Note: This is method is intended to import wav files from numpy.load method. So file names must be in .npy format.

  1. Set the correct NUMBER_OF_CLASSES on config.py to match desired label file to be exactly exported.
  2. Run wav_to_mfcc.py and wait to transform each:
python commands/wav_to_mfcc.py --src_path /home/vichoko/data/data/1d/svs-svd-bin-full --dest_path /home/vichoko/data/data/2d/svs-svd-N --src_label_prefix labels

Subset Data Folder

  1. Set the correct NUMBER_OF_CLASSES on config.py to match desired label file to be exactly exported.
  2. Run copy_data_by_label.py and wait to transform each:
python commands/copy_data_by_label.py --src_path /home/vichoko/data/data/1d/svs-svd-bin-full --dest_path /home/vichoko/data/data/1d/svs-svd-N --src_label_prefix labels

sync data folders through SSH

rsync -a /opt/file.zip [email protected]:/var/www/

rsync -avz -e "ssh -p $portNumber" user@remoteip:/path/to/files/ /local/path/