| Speech enhancement is a long standing problem in speech signal processing research [1, 2]. The distortion of speech signals with additive noise has hampered the success of numerous speech signal processing applications ranging from hearing aids [3, 4, 5], to speech coding [6] and automatic speech recognition [7, 4]. Up to now, most practical approaches to speech enhancement are waveform based, i.e. these approaches are (in principle) filtering the distorted signal to estimate the clean signal [8, 9]. In this thesis, we study how speech enhancement can be improved by restricting the class of targeted signals specifically to speech signals and employing parametric speech production models. The idea of fusing more speech specific information into the enhancement process is not entirely new [10, 11], but, so far, the performance of such methods has had only limited success.We present a new method for the enhancement of speech. The method is designed for scenarios in which targeted speaker enrollment as well as system training within the typical noise environment are feasible. The proposed procedure is fundamentally different from most conventional and state-of-the-art speech enhancement approaches. Instead of filtering a distorted signal we are re-synthesizing a new "clean" signal based on its likely characteristics. These characteristics are estimated from the distorted signal.To re-synthesize "clean" signal of high quality, a speech production model is needed. Throughout this thesis, we tried two different speech production models including autoregressive speech production model and inventory based speech production model. Successful implementations of the proposed methods are presented. Experiments show that the proposed method compares very favorably to other state-of-the-art systems in both objective and subjective speech quality assessments. Potential applications for the proposed method include vehicular speaker-phone communication systems and jet pilot communication systems. |
