Magnitude-only estimation of handset nonlinearity with application to speaker recognition

摘要: A method is described for estimating telephone handset nonlinearity by matching the spectral magnitude of distorted signal to output a nonlinear channel model, driven an undistorted reference. This "magnitude-only" representation allows model directly match unwanted speech formants that arise over channels and are potential source degradation in speaker recognition algorithms. As such, particularly suited algorithms use only information. The distortion consists memoryless polynomial sandwiched between two finite-length linear filters. Minimization mean-squared error, with respect parameters, relies on iterative estimation via gradient descent technique, using Jacobian correction term gradients calculated finite-element approximation. Initial work has demonstrated algorithm's usefulness reducing mismatch high- low-quality conditions.