We use cookies to ensure you get the best experience on our website
Почетна Преглед Помоћ News
Регистрација Пријавите се

Scheduled service maintenance on November 22

On Friday, November 22, 2024, between 06:00 CET and 18:00 CET, GIN services will undergo planned maintenance. Extended service interruptions should be expected. We will try to keep downtimes to a minimum, but recommend that users avoid critical tasks, large data uploads, or DOI requests during this time.

We apologize for any inconvenience.

mfrebo
/
align-vandam
1
0
Креирај огранак 0
Датотеке Дискусије 0 Захтеви за спајање 0 Вики

YODA repo to align vandam corpus using Montreal Forced Aligner.

Дрво: 1fd0ae8758
Гране Ознаке
align-vandam
Martin Frébourg 1fd0ae8758 edited comaprison.md пре 3 година
.datalad c11e500a15 [DATALAD] new dataset пре 3 година
.vscode 385198801d corrected with lucas' review пре 3 година
code dbd9efcf3f updated compare.py + confusion matrices пре 3 година
inputs 2cb46a39ea cleaned unnecessary files пре 3 година
outputs dbd9efcf3f updated compare.py + confusion matrices пре 3 година
.gitattributes c84fb1c33e Apply YODA dataset setup пре 3 година
.gitmodules 1d2b1741b5 [DATALAD] Recorded changes пре 3 година
CHANGELOG.md c84fb1c33e Apply YODA dataset setup пре 3 година
Comparison-summary.md 1fd0ae8758 edited comaprison.md пре 3 година
README.md 2820fd10b6 modified readme пре 3 година

README.md

Project

Dataset structure

  • All inputs (i.e. building blocks from other sources) are located in inputs/.
  • All custom code is located in code/.

Steps to generate aligned .csv from vandam-data .cha annotations

  1. Run code/csv2grid with annotations/cha/converted as input (converts the original .csv to .TextGrid)
  2. Run MFA Align with output files of previous step as input (with inputs/mfa-models/acoustic & inputs/mfa-models/dictionary as required)
  3. Run code/grid2csv to convert .TextGrids to .csv with outputs of previous step as input.

Steps for comparison of aligned segments with human annotator

  1. Use child-project sampler to generate 5x 1 minute segments (high-volubility) and outputs them in outputs/
  2. Use child-project eaf-builder with files generated at previous step and templates at inputs/eaf_templates
  3. Annotate segments by hand on ELAN
  4. Create csv dataframe with each segment in outputs/fivesegments-eaf
  5. Import that .csv with child-project import-annotations