340212 VU Speech Technologies (2024S)

Goals:

This course introduces basic techniques and methods of speech technology with a special focus on speech synthesis and speech recognition. Linguistic basics of speech production and basics of signal processing are also presented. Current areas of research as well as the current state of research are highlighted and discussed. Both classical methods, which are still relevant in hybrid architectures, and the latest methods based on neural networks will be presented.

Content:

14.3.:
Lecture 1
1. Introduction
2. Phonetics

11.4.:
Lecture 2
3. Signal Processing and classical vocoder
4. Minimum Edit Distance (MED) and Dynamic Time Warping (DTW)

18.4.:
Lecture 3
5. Hidden-Markov-models (HMM)
6. N-gram language models

25.4.:
Exercise 1

2.5.:
Lecture 4
7. Vector semantics and embeddings
8. Feed-forward Neural Networks (NN)

16.5.:
Lecture 5
9. Convolutional NN, RNN and LSTM
10. Transformer

23.5.:
Lecture 6
11. Speech synthesis: DNN based vocoders
12. Speech synthesis: DNN based acoustic models

6.6.:
Lecture 7
13. Speech recognition: DNN based acoustic models
14. Speech recognition: DNN based language models

13.6.:
Exercise 2

20.6.:
Programming exercise

Methodology:

Theoretical presentation of the basics of the field of language technology.
Development and implementation of a practical application to a current task in the field of the course.
Independent solving of exercises