An adaptive psychoacoustic filter for broadband noise reduction in audio signals

A novel adaptive noise reduction filter for audio signals is presented. The main aim of the method is to achieve a balanced tradeoff between noise suppression level and audible quality loss. The noise reduction is based on frequency domain processing of the low energy frequency bands. A perceptual model is applied to achieve the best quality judged by the listener. A major difference compared to conventional spectral subtraction methods, is the fact, that a prior estimate of the noise amplitude spectrum is not required in this method. This makes the proposed method well suitable for music recordings where the noise amplitude spectrum cannot be estimated.