Course details

Speech Signal Processing

ZRE Acad. year 2022/2023 Summer semester 5 credits

Current academic year

Applications of speech processing, digital processing of speech signals, production and perception of speech, introduction to phonetics, pre-processing and basic parameters of speech, linear-predictive model, cepstrum, fundamental frequency estimation, coding - time domain and vocoders, recognition - DTW and HMM, synthesis. Software and libraries for speech processing.

Guarantor

Course coordinator

Language of instruction

Czech, English

Completion

Examination (written)

Time span

  • 26 hrs lectures
  • 2 hrs exercises
  • 12 hrs pc labs
  • 12 hrs projects

Assessment points

  • 51 pts final exam (written part)
  • 14 pts mid-term test (written part)
  • 6 pts labs
  • 29 pts projects

Department

Lecturer

Instructor

Subject specific learning outcomes and competences

The students will get familiar with basic characteristics of speech signal in relation to production and hearing of speech by humans. They will understand basic algorithms of speech analysis common to many applications. They will be given an overview of applications (recognition, synthesis, coding) and be informed about practical aspects of speech algorithms implementation. The students will be able to design a simple system for speech processing (speech activity detector, recognizer of limited number of isolated words), including its implementation into application programs.

Learning objectives

To provide students with the knowledge of basic characteristics of speech signal in relation to production and hearing of speech by humans. To describe basic algorithms of speech analysis common to many applications. To give an overview of applications (recognition, synthesis, coding) and to inform about practical aspects of speech algorithms implementation.

Why is the course taught

Speech is the most common form of human communication and when we want to transmit information to others, we do mostly so by speaking. Automatic speech processing is necessary for processing of human-to-human communication (coding, or for example analysis of call center traffic), human-computer (voice command, voice search, authentication by voice) and computer-to-human (text to speech synthesis). The ZRE course will provide you with basic information on how speech is created, what usual features to represent it and what we do next. You will see that i tis an exciting combination of soft sciences", signal processing, machine learning, and other ingredients. Know-how acquired in ZRE is applicable also elsewhere - for example the problem of sequence recognition goes well beyond automatic speech recognition. The course is taught by members of the BUT Speech@FIT group, one of worlds prominent labs in speech data mining research.

Study literature

  • Gold, B., Morgan, N.: Speech and Audio Signal Processing, Wiley-Interscience; 2 edition.
  • Yu, D., Deng, L., Automatic speech recognition, Springer, 2016.
  • Rabiner, L. R., & Schafer, R. W. Theory and applications of digital speech processing, Pearson, 2011.
  • Psutka, J., Müller, L., Matoušek, J., & Radová, V., Mluvíme s počítačem česky, Academia, 2006.

Fundamental literature

  • Psutka, J.: Komunikace s počítačem mluvenou řečí. Academia, Praha, 1995, ISBN  80-200-0203-0 
  • Gold, B., Morgan, N.: Speech and Audio Signal Processing, John Wiley & Sons, 2000, ISBN 0-471-35154-7 

  • www stránka předmětu https://www.fit.vutbr.cz/study/courses/ZRE/public/

Syllabus of lectures

  1. Introduction, applications of speech processing. 
  2. Digital processing of speech signals.
  3. Speech production and its signal processing model. 
  4. Pre-processing and basic parameters of speech, cepstrum.
  5. Linear-predictive model. 
  6. Fundamental frequency estimation.
  7. Speech coding - basics
  8. CELP Speech coding. 
  9. Speech recognition - basics, DTW. 
  10. Hidden Markov models HMM. 
  11. Large vocabulary continuous speech recognition (LVCSR) systems. 
  12. Speaker and language recognition. Neural networks in speech processing. 
  13. Text to speech synthesis. 

Syllabus of numerical exercises

  1. Parameterization, DTW, HMM.

Syllabus of computer exercises

Except the last one, Matlab is used in labs.
  1. Introduction. 
  2. Linear prediction and vector quantization. 
  3. Fundamental frequency estimation and speech coding. 
  4. Basics of classification. 
  5. Recognition - Dynamic time Warping (DTW).
  6. Recognition - hidden Markov models (HTK).

Progress assessment

  • mid-term test 14 pts
  • project 29 pts
  • presentation of results in computer labs 6 pts

Course inclusion in study plans

Back to top