Skip to content

Latest commit

 

History

History
159 lines (127 loc) · 6.89 KB

submission_README.md

File metadata and controls

159 lines (127 loc) · 6.89 KB

Instructions to run the algorithm for mirex 2019 tasks

Audio Classical Composer Identification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package into the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 3GB
  • Time Taken ~ 10min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 10-15Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task classical
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 30min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task classical

Audio US Pop Music Genre Classification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package in to the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 10 GB
  • Time Taken ~ 30min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 25-40Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task us_pop
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 90min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task us_pop

Audio Latin Music Genre Classification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package in to the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 4GB
  • Time Taken ~ 10min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 10-15Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task latin
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 30min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task latin

Audio Music Mood Classification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package in to the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 1GB
  • Time Taken ~ 5min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 3-5Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task mood
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 10min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task mood

Audio K-POP Mood Classification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package in to the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 2GB
  • Time Taken ~ 5min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 5-8Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task kpop_mood
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 15min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task kpop_mood

Audio K-POP Genre Classification

Environment and setup

  • Download mirex.tar.gz. This is a conda package which has all the libraries required.
  • Create a folder for environment setup mkdir -p mirex2019_sm
  • Unpack the package in to the folder created tar -xzf mirex.tar.gz -C mirex2019_sm
  • Activate the conda environment in bash shell: source mirex2019_sm/bin/activate
  • Run the command conda-unpack
  • unzip the code directory

Feature extraction

  • Disk space requirements ~ 2GB
  • Time Taken ~ 10min
  • python extract_features.py --scratch {path_to_scratch_folder} --input_file {feature_extraction_list_file} --num_threads 4

Training

  • Disk space requirements ~ None
  • Time Taken ~ 7-10Hrs
  • python train.py --scratch {path_to_scratch_folder} --input_file {train_list_file} --num_threads 4 --task kpop_genre
  • If you get a Memory Error, please use --batch_size parameter to decrease the batch size to 16/8/4

Classification

  • Disk space requirements ~ None
  • Time Taken ~ 20min
  • python classify.py --scratch {path_to_scratch_folder} --input_file {test_list_file} --out_file {output_list_file} --num_threads 4 --task kpop_genre