Abstract
The goal of this research was to develop signal processing methods to improve the effectiveness of tactual vocoders. The proposed processing methods were designed to address the two major shortcomings identified in prior signal processing techniques used for tactile aids; (1) failure to transmit fundamental frequency information and (2) failure to track and code spectral information adequately, especially rapidly changing aspects of speech such as formant transitions. Three methods of processing were compared in experiments that measured the discriminability of tactile presentation of speech syllables. The first method was based on a traditional bandpass filtering approach. The second method was a based on homomorphic filtering to remove the glottal source components from the speech spectrum during voiced speech segments. The third method was based on a novel application of linear predictive coding (LPC) during voiced speech segments and traditional spectral signal processing methods during unvoiced speech segments. The homomorphic- and LPC-based methods were compared with and without peak picking post-processing algorithms. Overall results of these experiments show superiority for the LPC-based method and suggest that the LPC-based method may improve tactile perception of speech spoken at conversational rates.