Objective measures for hearing loss compensation techniques

Many techniques have been developed to compensate for hearing losses. With each new implementation a subject-based test was needed to verify its quality. The authors have developed objective measures to predict the results of a subject-based test. The measures can be computed directly from the original speech waveform and a speech waveform processed using the compensation technique. Certain parameters are extracted from the frequency spectra of the original speech waveform and used as the input to train a phoneme classifier. A feedforward neural network was used. Then, the same types of parameters were also measured from the processed speech waveform with some constraints of the hearing loss, such as hearing threshold incorporated. The objective measures can be used to predict the performance quality of new techniques being developed.<>