This paper describes a text independent continuous speech recognition system, which is built up from many steps. The method of the process is different in all steps, adapted to the particular problems of that level. The whole system is a combination of the application of neural nets and rule based processes, using segment based approach. The segmentation technics is language independent, and all the other processes can be adapted easily to other languages. The recognition works sentence by sentence. At the input the sound pressure is measured, and at the output the half-syllable series (first and second candidates) of the sentences are given. The recognition time of the sentences is about the pronunciation time of them, running the program on a PC 486.
Bibliographic reference. Vicsi, Klara / Vig, Attila (1995): "Text independent neural network/rule based hybrid, continuous speech recognition", In EUROSPEECH-1995, 2201-2204.