Lecturer(s)
|
-
Nouza Jan, prof. Ing. CSc.
|
Course content
|
Lectures 1. Speech - source of information, means of communication. Problems and challenges of speech recognition. 2. Principles and methods of speech signal parameterization 3. Cepstrum and cepstral features. 4. Hidden Markov Models (HMM). HTK platform. 5. HMMs applied for isolated word recognition. Viterbi decoder. 6. Phoneme based speech modelling and recognition. 7. Training of phoneme models, creation of training database. 8. Recognition of word sequences - modified Viterbi decoder. 9. Grammar based and stochastic language models. Language model training. 10. Hints for further improvement of speech recognition systems. 11. - 14. Work on individual or team project. Exercises 1. Recording of speech. Preparation of acoustic data for experiments. 2. Learning Hidden Markov Model ToolKit (HTK) 3. Speech parameterization in HTK 4. Training of whole-word models in HTK 5. Testing and experiment evaluation in HTK 6. Creation of training speech database 7. Speech recognition based on phonemes 8. Grammars 9. Connected word recognition 10. N-gram language models 11. -14. Work on individual or team project.
|
Learning activities and teaching methods
|
Monological explanation (lecture, presentation,briefing)
- Semestral paper
- 150 hours per semester
- Class attendance
- 56 hours per semester
|
Learning outcomes
|
This subject builds on basic knowledge acquired in Computer speech processing. It focuses mainly on explaining advanced algorithms of automatic speech recognition. Students will learn principles of probabilistic acoustic and language models used in recognition of isolated and continuous speech. In exercises, they will use software tools that allow them to train and test prototypical systems. At the end of the semester, they will work on a small project.
Student will get extended knowledge of modern speech recognition methods.
|
Prerequisites
|
Condition of registration: Exam from subject Computer speech processing.
|
Assessment methods and criteria
|
Combined examination
To get a credit active participation on excercises is required. Mark is based on the evaluation of the final project.
|
Recommended literature
|
-
Huang X., Acero A., Hon H.-W. Spoken Language Processing. A Guide to Theory, Algorithm and System Development. Prentice Hall. New Jersey, 2001.
-
Nouza J. (editor). Počítačové zpracování řeči (cíle, problémy, metody a aplikace). Technická univerzita v Liberci, 2001.
-
Psutka J. Komunikace s počítačem mluvenou řečí. Academia. Praha, 1995.
|