Parameter memory for speech synthesis

The paper describes digital formants speech coding method with parameters derived by homomorphic filtering from natural speech frames. The first five formants with the corresponding bandwidth and the value for pitch are represented in form of five byte lengths codewords in the parameter memory for the control of the speech production model to reproduce the desired frame. In order to realise good quality synthetic speech, with Romanian sounding and accent, our attention was focused to obtain suitable data for the parameter memory. For allophone synthesis the model and the memory becomes a single chip structure, which remains the basic module also for synthesis performed by concatenation of larger phonetic units, like demisyllables for example, with an external memory extension