ADSP2100-based speech synthesizer in Indian languages for vocally handicapped and spastic children

Presents the design, development and testing of a speech synthesizer, based on ADSP-2101. Frame-based analysis is carried out on sentences in Indian languages and the parameters obtained per frame are 10 reflection coefficients, pitch and the gain. These are used in a two-multiplier lattice model for the synthesis. Digitizing the speech signal at 6 kHz and at baud-rate of 2.3 kbits/s, the intelligibility of the synthesized sized sentence is reasonably good. The major implementational problem in the ADSP-2101 is the use of integer for the floating point numbers representing reflection coefficients and also the random data as source for unvoiced frame. This requires truncation of floating points up to a certain decimals. Intra-frame and inter-frame smoothing is required to improve the pronunciation. Vector quantization is also used to reduce the size of the storage (EPROM).