The Research Of Acoustic Feedback Control

The acoustic feedback problem widely exists in sound reinforcement and publicaddress systems. The acoustic feedback problem will make the system unstable, leadingto a bad acoustics or even damage to the audio sound equipment. If the impulseresponse order of the feedback path is known, the existing adaptive feedback controlalgorithms may be a good solution to the problem of the acoustic feedback. However, infact the impulse response order cannot be determined and changes with time, whichresults in the so-called "bias-order" problem. This thesis focuses on the "bias-order"problem and proposes an acoustic feedback control solution using audio sourceseparation.The expression of the "bias-order" in the stable solution of the adaptive filter hasbeen derived. Computer simulations and experiments show that the "bias-order"problem is harmful in acoustic feedback control and it’s difficult to be resolved by theexisting methods. So we try to apply the technique of audio source separation to theacoustic feedback control.A usual way for audio source separation is to project the mixture onto thetime-frequency space and derive the sources from the TF representation using a matrixdecomposition technique named non-negative matrix factorization. Sparse non-negativematrix factorization has been proven useful in source separation. In order to improve itsability in single-mixture audio source separation, we propose a complete sparenon-negative matrix factorization. Sparseness is constrained on both the temporal codematrix H and the spectral dictionary matrix W to get better overcomplete representations.Sparsity parameters are further adapted by using a variational Bayesian approach.Computer simulations and experiments have been conducted to verify the efficacy ofthe proposed single-mixture audio source separation method.Finally, we construct the model of the proposed acoustic feedback control method.Computer simulation results show that, whether in the hearing aids and public addresssystems, the proposed method can exhibit excellent performance in acoustic feedbackcontrol, and changes in the environment don’t deteriorate the performance.