- 121-Teaching and Learning through Visualised Speech Processing Experiments

In 1997 at Eurospeech conference we presented our education oriented system named VISPER. Its goal is to demonstrate in visual form the basic concepts and algorithms used for isolated-word recognition (speech endpoint detection, signal analysis, parameterization, DTW and HMM classification techniques). The system environment gives a student a lot of freedom in setting up and running his own experiments. He can define his own vocabulary, make own recordings, select arbitrary feature subsets and modify the classification or training procedures. The experiments are visualised so that the student can observe not only results but also all the relevant steps in the recognition process. At our university the VISPER has been used with success in MSc and PhD courses in speech processing.