(2) Test the existing baseline ASR model [2]

[PanosAntoniadis]

• Evaluate the model using Kaggle dataset.
• Evaluate the model using Journalism dataset.

[PanosAntoniadis]

The kaggle dataset in the required form for evaluation by Sphinx can be found here: https://drive.google.com/open?id=18hErfhgmXmhJ0VraVX4zb4SJ2UOOZl8l.

Results using Sphinx aligner https://cmusphinx.github.io/wiki/tutorialtuning/ :
TOTAL Words: 37138 Correct: 17879 Errors: 21304
TOTAL Percent correct = 48.14% Error = 57.36% Accuracy = 42.64%
TOTAL Insertions: 2045 Deletions: 2362 Substitutions: 16897

[PanosAntoniadis]

The Journalism dataset in the required form for evaluation by Sphinx can be found here:
https://www.dropbox.com/sh/a8dkcgchb3cxgnc/AAA-7uxX8embvJWPOW-yQFTGa?dl=0

Results using Sphinx aligner https://cmusphinx.github.io/wiki/tutorialtuning/ :
TOTAL Words: 11532 Correct: 5038 Errors: 6811
TOTAL Percent correct = 43.69% Error = 59.06% Accuracy = 40.94%
TOTAL Insertions: 317 Deletions: 936 Substitutions: 5558

@asymeon