Signal processing for a visual aid to improve speech intelligibility of the hearing impaired

Many people with hearing loss have trouble perceiving and understanding speech—particularly in noisy environments. The goal of this research is to evaluate signal processing strategies for use in a visual aid. This aid would present speech information through peripheral vision using LEDs mounted in the frame of a pair of glasses.; Two strategies for detecting unvoiced fricative and affricate phonemes based on high frequency content were evaluated. The High Frequency Energy (HFE) method calculates the energy in the speech signal after high pass filtering. The High Frequency Energy Ratio (HFER) method compares the energies in a high and low frequency band. The effect of parameter variation on the receiver operator characteristic (ROC) of each method was investigated. From the results, we have found parameter settings that maximize detector performance. The maximum average discrimination rates were 89.6% for the HFE method and 91.6% for the HFER method.