Evaluation script for forced aligned textgrid

• This script evaluates a TextGrid (i.e. prediction; usually of forced alignment) with regard to one reference TextGrid (i.e. target; usually human segmented).

• Two TextGrid files of an audio file:
  • One by automatic forced alignment (FA)
  • One by human as a reference

• 1. Name of .TextGrid file to evaluate (usually a forced alignment)
• 2. Name of .TextGrid file to refer to (usually by human)
• 3. Tolerance (ms) for accuracy calculation (default: 20 ms)

• evaluation.csv
  • includes 5 columns
    • First 3 columns:
      • records of 3 given input arguments
    • Last 2 columns:
      • 1. Total Error (ms)
      • 2. Alignment Accuracy (%)
           = what percentage of the automatically labeled boundaries are within a given time tolerance (threshold) of the manually labeled boundaries.

• Evaluate performance of a forced aligned TextGrid, referencing to human alignment:

    $ python evaluateFA.py <FA_textgrid> <human_textgrid> <tolerance>
  

• For example:

    $ python evaluateFA.py fa.TextGrid human.TextGrid 30