Speech Synthesis by Rule: An Acoustic Domain Approach
Lawrence R. Rabiner · Bell System Technical Journal · 1968
A new approach to speech synthesis by rule has been formulated and evaluated. A discrete set of symbols (phonemes and stress marks) is converted to a continuous acoustic waveform by a two-step transformation. The first step involves conversion from phonemes to control signals capable of driving a terminal analog speech synthesizer. The second step is conversion from control signals to the acoustic waveform. This paper presents a design for the terminal analog synthesizer and discusses the new features of this device. It discusses in detail the method of converting from phonemes to control signals. It places primary emphasis on determining the formant frequency control signals and the fundamental frequency contour, and presents models for determining these contours from the input data. The paper includes an experimental evaluation of the entire technique in terms of word intelligibility scores and consonant confusion matrices.