AE2M31ZRE - Speech processing
AE2M31RAT - Speech technology in telecommunications

Schedule of Seminars
Summer semester 2016/2017

Tools for practical exercises:

  1. Tu - 21.2.2017 (Pollák) - Introduction: speech signals, analysis tools
    - interactive work with signals in the environment of MATLAB, Praat, Wavesurfer
    - recording of database for further work during the semester
    Guidelines for the seminar

  2. Tu - 28.2.2017 (Pollák) - Basic time- and frequency-domain characteristics
    - Energy, power, intensity, RMS value
    - Zero-crossing rate
    - Spectrogram and filter-bank spectrum
    Guidelines for the seminar

  3. Tu - 7.3.2017 (Pollak) - Pitch estimation
    - Pitch estimation from autocorrelation function
    - Necessary pre-processing and post-processing
    - Estimation within MATLAB and Praat (Wavesurfer)
    Guidelines for the seminar

  4. Tu - 14.3.2017 (Fiala) - Cepstrum, cepstral distance and voice activity detection
    - DFT, LPC, and MFCC cepstrum
    - Cepstrum of longer speech utterance
    - Energy-based and cepstral based voice activity detection
    Guidelines for the seminar

  5. Tu - 21.3.2017 (Pollák) - LPC spectrum and formant estimation
    - computation of LPC spectrum (spectrogram)
    - formant estimation from AR model (LPC parameters)
    - Estimation within MATLAB and Praat (Wavesurfer)
    Guidelines for the seminar

  6. Tu - 28.3.2017 (Pollák) - LPC vocoder: MATLAB implementation of principles parts
    - AR model of speech signal
    - elementary vocoder with error signal exication
    - quantization within error-excited vocoder (RELP simulation)
    - vocoder with atificially generated exication
    Guidelines for the seminar

  7. Tu - 4.4.2017 (Pollák) - Basics of classification based on GMM and VQ
    - principles of GMM and VQ
    - the recognition of vowels based on formants or cepstrum
    Guidelines for the seminar

  8. Tu 11.4.2017 (Pollák) - Speaker verification on the basis of vector quantization
    - cepstrum distribution for given speaker, differences among speakers
    - creation of speaker codebook
    - realization cross verification from available database
    - interactive on-line verification
    Guidelines for the seminar

  9. Tu 18.4.2017 (Pollák) - Speaker identification on the basis of GMM
    - GMM based speaker verification (a simple modification of the task from last seminar)
    - usage of extended feature vector
    - search of selected speaker within a set of unknown utterances
    Guidelines for the seminar

  10. Tu 25.4.2017 (Petr Pollák) - DTW-based speech recognition
    - computation of distance matrix
    - looking for the optimal path through given distance matrix
    - simple recognizer of isolated digits
    Guidelines for the seminar

  11. Tu - 2.5.2017 - Cancelled - the change of scheduling

  12. Tu - 9.5.2017 (Pollák) - HMM-based recognition I
    - Recognition of vowel sequence
    - Computation of GMM for particular states
    - Emitted probability in selected states
    Guidelines for the seminar

  13. Tu - 16.5.2017 (Pollák) - HMM-based recognition II
    - Emitted probability in all states
    - Transient probabilities
    - Passing through HMM, Viterbi decoding, optimum path
    Guidelines for the seminar

  14. Tu - 23.5.2017 (Pollák) - HMM-based recognition III
    - Backward Viterbi decoding
    - Occupacy likelihood of particular HMM states
    Guidelines for the seminar