An Automatic Method to Develop Music with Music Segment and Long Short Term Memory for Tinnitus Music Therapy

Abstract

Tinnitus is a perception of sound when no external sound is present. It has seriously affected patients' life. Music is an option to relieve tinnitus in clinic, as it can bring enjoyment to listeners. However, existing music used in tinnitus therapies has limited duration, it is usually repetitively played during the long-term treatment and may not be helpful for relaxation. Moreover, individualized preferences of patients are ignored in most cases. Both of them may hinder tinnitus relief. Although existing methods can synthesize specific music that has unlimited duration and is not repetitively played, the synthesized music has defects with pitch mutations and long pitch durations. Moreover, characteristics of these synthesized music have not been confirmed by tinnitus patients. Therefore, this study presents an automatic method to develop the specific music based on music segments from existing music and long short term memory (LSTM). Numerical results indicate that specific music developed in this study not only retains characteristics of original music, but also overcome the defects above. Besides, a total of 30 tinnitus patients and 10 tinnitus-free volunteers participated the auditory experiment. Auditory results are consistent with numerical results and also suggest that tinnitus patients can perceive feelings that are conducive to tinnitus relief after listening preferred music. Therefore, the developed music presents a possible complement to tinnitus treatment in clinic.