Experimental study on enhancement method for high-frequency hearing loss in Chinese digital hearing aids

A nonlinear frequency lowering algorithm based on the sinusoidal speech model is proposed to improve high-frequency audibility. In a sinusoidal model, the speech waveform can be characterized by the amplitudes, frequencies, and phases of the component sine waves. These parameters are main processing objects in the study. The frequency spectrum is split into six parts according to octaves to avoid spectral distortions in the sounds. For the part which is below and closest to a threshold frequency of the patient, only the amplitudes amplification method is used. Frequencies above a threshold frequency are compressed by different ratios according to the speech intelligibility in different frequency ranges and shifted to the remaining regions below the threshold frequency. Phases of the transformed frequency are adjusted to the same as the phases of the low-frequency closest to this frequency. In this study, ten subjects were tested. Results in the speech recognition experiment show the mean speech intelligibility is improved with at least 45% after the training of five weeks. In future, the number of subjects should be increased. Furthermore, hearing loss characteristics of different patients should be carefully analysised to design a more reasonable and more detailed frequency lowering algorithm.