Audio Denoising Using Wavelet Filter Banks Aimed at Real-Time Application
Peter Kassakian · DSpace@MIT (Massachusetts Institute of Technology) · 1999
In this thesis we implement an audio denoising system targeted at real-time application. We investigate the delays associated with wavelet filter bank systems and propose methods to overcome them. We consider computational delays as well as structural delays and deal with each separately. We find that good denoising performance is achieved when filters are chosen so as to transform the input into an appropriately scaled time-frequency domain. Multiplications required for four different implementations of the system are compared and we conclude that a scheme that performs multiplications as far downstream as possible proves to be the best of the four options. Finally we propose several topics for further research including investigations into the human auditory system as well as studies of creative filter bank topologies.