This project contains materials for creating a Russian language acoustic model:
-
Guidelines.pdf contains full guide about work process and description of project files (in Russian only);
-
/ru_base contains language models (8 hours of speech), dictionary, phonemes, parameteres of training and training data;
-
/scripts contains utility scripts which are not really necessary;
-
/theory contains some must know information about linux and speech recongnition fundamentals.
ru_base/etc/ru_base_large.lm and ru_base/lm_train_data.txt are compressed due to github size restrictions.
- CI, 8 DEN:
TOTAL Words: 197 Correct: 114 Errors: 94
TOTAL Percent correct = 57.87% Error = 47.72% Accuracy = 52.28%
TOTAL Insertions: 11 Deletions: 5 Substitutions: 78
- CD, 16 DEN, 2000 sen + LW tune
TOTAL Words: 197 Correct: 142 Errors: 81
TOTAL Percent correct = 72.08% Error = 41.12% Accuracy = 58.88%
TOTAL Insertions: 26 Deletions: 2 Substitutions: 53
- CD, 16 DEN, 2000 sen + lm_test
TOTAL Words: 197 Correct: 194 Errors: 5
TOTAL Percent correct = 98.48% Error = 2.54% Accuracy = 97.46%
TOTAL Insertions: 2 Deletions: 0 Substitutions: 3