Phonetic Recognition Experiments with Recurrent Neural Networks

In order to prove the potential power of "learning by examples" paradigm for problems of Automatic Speech Recognition, an experiment was set up, regarding an extremely difficult Italian phonetic recognition problem: the automatic discrimination of the so called Italian alphabet i-set: /1311, /tSi/, /di/, /dZi/, ill, /pi/, /ti/, lvi/ plus other two i-like stimuli /LI/, /si/. The achieved speaker independent mean recognition rate was around 65%, but either better results are expected at the end of the experiments going on.