| 摘要 |
The 128 words of a given list are stored for Vocoder-type speech synthesis as pairs of purpose-chosen "syllables", one possibly mute, (seven-teen, cf. seven-(silent) of purpose-chosen "phonemes" (s-ev-n, vol-t-s). Each phoneme waveform (118) has its time derivative digitalised at 4-bit accuracy and modified to simulate a standard vocal pitch (120), then simplified, when possible, to the iteration of an initial interval (122) and/or the alternation of a significant and a negligible interval (124). The simplifications are stored in code and the initial/significant intervals of the derivative approximated in a 'smoothed', limited-gradient curve, which is stored as a set of 2-bit parameters for its gradients. In the synthetic reproduction, the stored words can be addressed singly and/or combined into a limited number of chosen phases. |
