Talk to me


Erstellt von


Author of this exhibit: IMAGINARY. It is based on research by Andreas Krug, M. Sc., and Jens Johannsmeier, M. Sc., of the Artificial Intelligence Lab (Otto-von-Guericke Universität Magdeburg) under the supervision of Prof. Dr.-Ing. Sebastian Stober.


Unterstützt durch

Carl Zeiss Foundation

An automatic speech recognition exhibit.
Push the “Record”-button and speak into the microphone. You can speak in German or English by selecting the corresponding language before recording.

It is normal for us to use the speech recognition of our mobile phones when we want to find out the fastest way home. But actually, recognizing our language correctly, is not an easy task for a computer. When we speak, we blend multiple letters to a sound or make no pauses between the words. Each one of us also speaks differently due to our own accent or dialect. “Talk to me” shows us the processes necessary in the background for a machine to recognize our speech. Its neural network was trained with hundreds of hours of voice recordings in English and German in order to be able to detect every letter as correctly as possible.