Algorithm selection and software time/Space optimization for a DSP micro-based speech processor for a multi-electrode cochlear implant

Selection of appropriate algorithms for pitch and formant extraction, in conjunction with use of innovative programming techniques, enables these speech-related features to be derived in real-time on a low-power TMS 320C10 based-system using about 60% of that chip´s computing power. Moreover, since LPC-based techniques are used, an accurate low-bit-rate model of the vocal system is inherent in the raw parameters extracted. This programmable system thus provides a flexible foundation for exploring the utility of other fundamental speech features in addition to pitch and formants for excitation of a cochlear prosthesis.