/caqe_segmentation

Code repository for "Re-visiting the Music Segmentation Problem with Crowdsourcing"

Primary LanguageJupyter Notebook

caqe_segmentation

This is the experiment and plotting code repository for the paper, ''Re-visiting the Music Segmentation Problem with Crowdsourcing'', accepted to ISMIR2017 at Suzhou, China.

The AMT-pilot dataset

The annotaions gathered in this work by Cheng-i Wang should be referred to as the ''AMT-pilot'' database. This dataset contains AMT annotations on 8 songs from Beatles-TUT and SALAMI.

Code Summaries

caqe_boundary python module

A flask-sqlalchemy setup that builds databases specifically for analyzing annotation data.

Jupyter Notebooks

  • beatles_db_extract.ipynb and salami_db_extract.ipynb extract AMT annotations from beatles_accpted.csv and salami_accepted.csv and build SQL databases.
  • beatles_stats.ipynb and salami_stats.ipynb plot graphs and calculate statistics used in the submitted paper.

CSV Files

beatles_accpted.csv and salami_accepted.csv contain the extracted accepted annotations from AMT using the CAQE extension developped for this paper

Audio Folders

Contain original song files, chopped 20-seconds clips and randomized 10-clip tasks.

Misc

.dump files are the databases for the entire AMT tasks. Utility jupyter notebooks in mturk_utils folder are used to chopping songs and assigning chopped clips to tasks.