Digital Signal Processor: Speech Synthesis

Speech synthesis is one of several promising areas of application for the digital signal processor (DSP) chip described in this issue of the Bell System Technical Journal. Its computational power, low cost, and easy interfacing are the properties that allow a design of a standalone speech synthesizer with very few components outside the DSP chip. Such a synthesizer can be used in a variety of devices intended for providing new services in a business environment, as well as in future residential services. One of the most successful ways to synthesize speech is based on a linear predictive coding (LPC) model of a vocal tract. See Refs. 1, 2, and 3. In this model, the vocal tract is approximated by a linear dynamic system driven by impulse sequences for voiced sounds, or by white noise for unvoiced sounds (Fig. la). The difference equation for such a representation is 1621 EXCITATION N 1-1*;*-' /=1 SPEECH SOUNDS WHITE NOISE IMPULSE TRAIN (a) (b) Fig. la--Polynomial form of transfer function. Fig. lb--Lattice form of transfer function. i=N