Method and Tool Innovations for Speech Science Education (M.A.T.I.S.S.E.) April 16-17, 1999 London, UK

The Set of Exercises in Digital Speech Processing

Jan Uhlir

Czech Technical University, Prague, Czech Republic

Engineers who study electronics and communications technology are frequently confronted with algorithms and techniques concerning speech signals. We have prepared the study course, which covers fundamentals of phonetics and speech processing. In the first lectures, students learn the basics of speech communication. Auditory and articulation models are described. The next theme is devoted to the speech processing techniques. Typical parameters of the signal are introduced. The principles of text to speech synthesis are presented in the third part of course. Finally, automatic speech recognition and its algorithms are studied in the last part of course.

The experimental works, which follow the lectures, will be presented in the contribution. Experiments are made in Matlab. We have worked out the set of M-files, which guides the student through the main topics of lectures. All M-files are prepared as simple and open programs. Students can change constants and modify the calculations. The results are visualised and in proper cases demonstrated by audio output.

Until now, the set of M-files covers following topics:

• Quantization, quantization noise, nonlinear quantizers, DPCM, sigma-delta quantizer
• Predictive quantizer, short term parameters (LPC)
• Short term spectra, FFT, LPC, cepstra
• Formant structure of phones, voiced vs. unvoiced, pitch period
• Spectral distance measure, DTW algorithms

Full Paper

Bibliographic reference.  Uhlir, Jan (1999): "The set of exercises in digital speech processing ", In MATISSE-1999, 53-56.