Audio DSP: Time and Frequency Varying Gain Compensation for Non-Optimal Listening Levels

Although the human hearing system is very complex, several models exist that explain parts of the hearing system. This thesis uses one of these models, the contours of equal loudness, to make music played at low listening levels sound more like it does at the intended listening level. The perceived frequency balance of music varies with the listening level. This is especially noticeable at low listening levels, where frequencies below 500Hz seem attenuated. Moreover, hearing perception exhibits non-linear dynamic range compression, most evident at low frequencies. A system is designed where filter banks and power measurements estimate the time-varying power of low frequency parts of the audio signal. The time-varying power of each narrow frequency band is compared to the contours of equal loudness, and changes made to get the same frequency balance as at the intended listening level. The thesis covers the design, implementation and performance of this system