Speech Processing - B2M31ZRE

Credits 6
Semesters Summer
Completion Assessment + Examination
Language of teaching Czech
Extent of teaching 2P+2C
Annotation
The subject is devoted to basis of speech processing addressed to students of master program. Discussed speech technology is currently applied in many systems in different fields (e.g. information dialogue systems, voice controlled devices, dictation systems or transcription of audio-video recordings, support for language teaching, etc.). Students will learn basic algorithms for speech analysis (spectral analysis, LPC, cepstral analysis, pitch, formants, etc.), principles of speech recognition (GMM-HMM, ANN-HMM systems, small and large vocabulary recognizers), speaker recognition (based on VQ and GMM), speech synthesis or speech enhancement. Further information can be found at http://noel.feld.cvut.cz/vyu/ae2m31zre. Pro zapsané studenty jsou detailní informace na výukovém portálu Moodle FEL.
Study targets
The goals of the subject is to introduce used speech technology in the most important multimedia applications. Students should manage the knowledge as basic characteristics of speech signal, speech enhancement, speech recognition, speech synthesis, audio-visual speech processing, etc. Students will practice basic tasks of speech processing in MATLAB environment and also other publicly available tools for speech analysis will be used. 
Course outlines
1. Introduction - speech production and perception model, basic characteristics (phonetic and articulatory)
2. Spectral characteristics of speech signal (DFT and LPC spectrum)
3. Cepstral reprezentation of speech. Recognition features. Voice Activity Detection.
4. Speech enahncement (additive and convolution noise, one-channel and multi-channel systems)
5. Basic classification approaches and techniques (GMM, HMM, VQ, ANN, DNN)
6. Speaker verification and identification. Language recognition.
7. Small and large vocabulary speech recognition (DTW, GMM-HMM, LVCSR, HTK and KALDI tools).
8. Modern LVCSR systems (DNN-HMM). Adaptation techniques. Advanced speech features.
9. Speech synthesis - basic principles (concatenative and formant synthesis, PSOLA)
10. Audio-visual speech recognition
11. Speech coding.
12. Hearing aids and cochlear implants (anatomy and hearing model, speech processing)
13. Multimedia systems with voice input (dialog systems, logopaedy, language teaching)
14. Databases for speech technology systems. Reserve.
Exercises outlines
1. Introduction: speech signal, tools for analysis, sources of speech signals
2. Basic time-domain and spectral characteristics
3. Fundamental frequency (pitch) estimaton
4. LPC spectrum and formant estimation
5. Cepstrum and cepstral distance: voice activity detection.
6. Basic classification techniques (GMM, VQ, HMM): vowel classification
7. Speaker verification based on VQ
8. Speaker identification based on GMM
9. DTW based recognition: simple recognizer of particular words
10. HMM based recognition: basic tasks and demonstration of HMM modelling
11. Suppression of additive noise in speech signal
12. Convolutory noise suppression
13. Speech synthesis: implementation of formant synthesis, demonstration of available tools
14. Reserve. Credits
Literature
[1] Huang, X. - Acero, A. - Hon, H.-W.: Spoken Language Processing. Prentice Hall 2001.
Requirements
Bases of digital signal processing are supposed as preliminary knowledge.