Theoretical and practical aspects of computer speech processing

Abstract

The design process of computer speech systems is presented. Going over the solutions of neural networks, the systolic structure gives the biggest possibility for smart realization of a speech processor. A speech processing system, built generally as software surroundings, occupies an immense memory space. For real-time application it is necessary to combine the existing algorithms. Further increasing of the speed can be also achieved by using hardware-software co-design. The approach presented in this paper, permits to design such effective systems of human voice synthesis and speech identification. Since the spoken language can be very often ambiguous, then it was useful to include a fuzzy logic approach and also artificial neural networks, to obtain more effective methods of speech processing. Mixed analog-digital realizations are also discussed. Practical realizations of the current mode building blocks ensure the cheapest realization of CMOS integrated circuits.