Digital Speech Processing

Course: Digital Speech Processing

Code: 3ФЕИТ05005

ECTS points: 6 ECTS

Number of classes per week: 3+0+0+3

Lecturer: Asst. Prof. Dr. Branislav Gerazov

Course Goals (acquired competencies): The goal of the course program is to allow students to acquire a wide knowledge of the techniques for the analysis, synthesis and recognition of speech signals. It is designed to bring close the various approaches and applications of digital speech processing through studying the state-of-the-art.

Course Syllabus:

1. Fundamentals of digital audio, principles of digitisation, oversampling, jitter

2. Working with audio signals in the digital domain; quantisation, dither, noise shaping.

3. Fourier transform, Z-transform, amplitude and phase spectrum.

4. Sliding window method, short time Fourier transform (STFT), spectrograms.

5. Fundamentals of digital filters, filtering, FIR, IIR, FIR filter design.

6. LP, HP, BP, BS, and Notch filters, filterbanks, equalisation.

7. Basics of speech production, source filter model, LP analysis, VOCODER; compression of speech LP10, CELP.      8. Machine learning for speech signals, basics of ASR, feature extraction, MFCCs.

9. DTW, HMM, GММ.

10. Deep learning in ASR, NN, DNN, RNN, LSTM, CNN.

11. Speaker recognition, GММ, ЕМ, UBM, LLRM.

12. Speech synthesis, concatenative, articulatory and formant synthesis, parametric synthesis with HMM and NN.

13. Deep learning for speech synthesis Wave-Net.

Literature: