 |
Method and Tool Innovations for Speech Science Education (M.A.T.I.S.S.E.)April 16-17, 1999 |
 |
|
Method and Tool Innovations for Speech Science Education (M.A.T.I.S.S.E.)April 16-17, 1999 |
|
At Eurospeech’97 we presented system VISPER that had been developed as a teaching and learning tool for introductory courses in speech science and technology. The system is aimed at explaining basic speech recognition paradigms and procedures like, for example, signal processing, feature extraction, word distance measurement, hidden Markov modelling. Learning and understanding these topics is supported by VISPER’s graphic environment that allows for visualisation and animation of the essential procedures. A large choice of options and settings makes the system an educational workbench enabling students to learn from experiments.
In the paper we describe our experience with using the VISPER in university (MSc and PhD) courses in speech processing. We show that the scope of the experiments provided by the system is quite large, starting from a simple DTW based speaker-dependent word recognition task and ending with, e.g., a real-time continuous HMM speaker-independent classifier. After several introductory lectures and seminars the students may start their individual investigations. Is one feature (e.g. energy) and one template good enough for a 5-word recogniser? How does a DTW path look like for a pair of same words and different words? How is a difference between single-mixture and multi-mixture continuous HMMs? These are typical questions that the students ask. In our courses they try to find the answers by themselves. Simply by starting the VISPER, entering their own vocabularies, recording their own data, selecting arbitrary feature subsets, choosing algorithms and running their own visualised and animated experiments.
Bibliographic reference. Nouza, Jan (1999): "Teaching and Learning through Visualised Speech Processing Experiments ", In MATISSE-1999, 121-124.