1 Acoustic Technology, Department of Electrical Engineering, Technical University of Denmark2 Technical University of Denmark
A new method for the objective estimation of sound quality for both normal-hearing and hearing-impaired listeners has been presented: OSSQAR (Objective Scaling of Sound Quality and Reproduction). OSSQAR is based on three main parts, which have been carried out and documented separately: 1) Subjective sound quality ratings of clean and distorted speech and music signals, by normal-hearing and hearing-impaired listeners, to provide reference data, 2) An auditory model of the ear, including the effects of hearing loss, based on existing psychoacoustic knowledge, coupled to 3) An artificial neural network, which was trained to predict the sound quality ratings. OSSQAR predicts the perceived sound quality on two independent perceptual rating scales: Clearness and Sharpness. These two scales were shown to be the most relevant for assessment of sound quality, and they were interpreted the same way by both normal-hearing and hearing-impaired listeners. The scales were found not to be absolute, thus OSSQAR cannot predict the absolute sound quality, but it can be used to rank the sound quality. Using test data from the present subjective rating experiment, the prediction error of OSSQAR was found to be only slightly larger than the random variance in the subjective ratings. Analysis of the neural network after training could not identify qualitative relations between the physical signal parameters and the perceived sound quality. Further verification of OSSQAR with new signals and distortion types will be required to assess how general and reliable OSSQAR is, and to identify the limitations of its application.