snippets for MIREX AME task

• Salamon, melodia
• Durrieu, separateLeadStereo
• Bosch, SourceFilterContoursMelody
• Kum, MelodyExtraction_MCDNN
• Bittner, ismir2017-deepsalience
• Hsieh, Melody-extraction-with-melodic-segnet
• Kum, melodyExtraction_JDC
• Basaran, ismir2018_dominant_melody_estimation
• Lu, Vocal-Melody-Extraction

• MedleyDB (vocal/instrumental)
• RWC Popular, RWC Royalty free
• ADC 2004 (vocal/instrumental)
• Orchset (instrumental)
• Mirex-05 (vocal/instrumental)
• iKala

Except for the requirements needed by the algorithms repos above, the code use Python 3.6 or higher version, and requirements is listed in requirements.txt. You could install the requirements with command pip install -r requirements.txt.

Since this repo does not include any dataset files and algorithms repos, you should have download them independently and configure the path to the datasets and algorithms in the configs/configs.py.

train the proposed model.

Usage: train.py [OPTIONS]

Options:
  --resume BOOLEAN  resume training state from .pkl file
  --debug BOOLEAN   use small dataset to debug faster
  --help            Show this message and exit.

inference using the trained model. Pretrained models can be downloaded from google drive, download and put them in the data/ folder.

Usage: predict.py [OPTIONS] AUDIOFILE [MELFILE]

Options:
  --model TEXT   pretrained model path
  --cpu BOOLEAN  use when have no cuda support
  --help         Show this message and exit.

example: python predict.py <path-to-input-audio-file> <path-of-output-csv-file>

evaluate state-of-the-art algorithms on various datasets.

Usage: evalAlgos.py [OPTIONS]

Options:
  --force BOOLEAN   overwrite evaluation results
  --dataset TEXT    using specific dataset
  --algorithm TEXT  using specific algorithm
  --help            Show this message and exit.

a convenient command line interface for calling algorithms.

Usage: algorithmsCLI.py [OPTIONS] ALGO AUDIOFILE [MELFILE]

Options:
  --help  Show this message and exit.