登入選單
返回Google圖書搜尋
Single and Dual Microphone Speech Enhancement Techniques for Hearing Devices
註釋Speech Enhancement (SE) systems form a vital front end in many applications. In the last two decades, extensive research has been carried out in single and multi-microphone SE techniques. In this thesis, four novel SE techniques have been proposed. The first SE method is used for Quasi-Periodic noise environment. It exploits the knowledge of period in the noisy part to increase the Signal to Noise Ratio. The second method uses Supervised Machine Learning approach to estimate the Wiener mask to suppress the noisy speech using single microphone and works well for quasi periodic noise. The third method is a Blind Source Separation (BSS) technique based on Independent Component Analysis (ICA) for the underdetermined case and instantaneous mixture. The fourth method is a BSS technique which uses Independent Vector Analysis (IVA) for convolutive mixtures. Objective and subjective evaluation of the developed methods show drastic improvements in speech quality and intelligibility.