Speech enhancement system based on lattice filter and system identification

We have investigated a noise reconstruction system (NRS) with an adaptive line enhancer (ALE) as a noise reduction system. The NRS uses a linear prediction error filter (LPEF) and a noise reconstruction filter (NRF). The NRS assumes that the background noise is generated by exciting a linear system with a white noise. First, the white signal is estimated by the ALE and the LPEF. Next, the background noise is reconstructed from the white noise by estimating the liner system by the NRF. The conventional NRS uses a transversal filter as the LPEF. However, the speech component remains in the estimated white noise. As a result, the quality of enhanced speech is degraded. In order to solve the problem, we introduce a lattice filter as a LPEF. Since the structure of a lattice filter approximates a vocal-tract filter for the speech production process, the lattice type LPEF can estimate the vocal-tract parameter accurately than a transversal type LPEF. Therefore, the ability of a speech enhancement is improved.