Speech synthesis from phonemic transcription

In this paper we will describe the portion of the text-to-speech conversion system which accepts a phonetic transcription, stress, and timing information as its input, and outputs the corresponding speech wave. The synthesis program is a form of dyadic concatenation of LPC area segments obtained from natural speech; in addition, the program also calculates the amplitude and intonation. The dyadic concatenation is performed using a matrix of phoneme transitions to obtain the LPC area parameters; the amplitude is obtained by rules which depend on the class of the phonemes involved in the amplitude computation. The fundamental frequency contour is obtained by an adjustment of a stored fundamental frequency contour to fit the utterance to be synthesized. This adjustment is performed by selecting the portion of the stored fundamental frequency contour which fits the number of words and their function in the utterance. The location of the peak of the contour for each word in the utterance is then adjusted to fall at the location of the primary stress.