Digital signal processor-based feature extraction vocoder for the deaf

Abstract

An innovative feature detection tactile vocoder based on Linear Predictive Coding (LPC) processing algorithms is described. The device utilizes the 80386 microprocessor and the TMS32OC30 floating point digital signal processor. Speech processing is implemented in real-time. The vocoder algorithms allow the detection spectral peaks corresponding to the first 3 formants of the vocal tract as well as fundamental frequency. This LPC approach enhances the ability of vocoders to track formant transitiotu, fundamental frequency and minimizes spatiotemporal masking.

Evaluation of an LPC-based tactile aid for the deaf

“…A floating point DSP was used to perform LPC analysis and fundamental frequency estimation on voiced segments, FFT analysis on unvoiced segments and calculation of vibrator auenuation values [3].…”