適用於雙耳助聽器之二元遮罩噪音消除演算法

Abstract

在現實環境中，在我們身邊通常有好幾個噪音源。對於配戴助聽器的聽損者，噪音是一個令人感到不舒服的問題。要克服這個問題，現代數位助聽器都會具備降噪功能。過去有許多降噪演算法被提出，他們主要在改善語音品質，也就是所謂的信噪比。然而，一些研究者實際進行聽力實驗，結果顯示降噪並不總是提高語音清晰度。在某些情況下，它會傷害語音理解度，甚至比原本嘈雜語音更糟。 在這篇論文中，我們提出一個基於二進制遮罩的雙耳助聽器的降噪設計。它可以保留對人類雙耳聽覺非常重要的聲音定位線索。此外，它採用了ideal binary mask (IdBM)的概念以提高語音理解度。本論文提出的雙耳降噪算法，不僅像以前的降噪提高語音品質也提高語音理解度。對於聲音定位，它對輸入與輸出訊號保持相同的雙耳聲音差異。從模擬結果，講話的品質（信噪比改善）不比其他的降噪演算法差，此外，也增強了語音理解度。

In the real environment, there are usually several noise sources around us. For the patient wears the hearing aids, the background noise is an annoyed problem. To conquer this problem, the function of noise reduction is equipped in the modern digital hearing aids. There were many noise reduction algorithms developed in the past decades which improved the speech quality in the evaluation of SNR. However, some researcher took the listening experiment and indicated that the noise reduction algorithm did not always improve the speech intelligibility. In some case, it would hurt the speech intelligibility even worse than the original noisy speech. In this thesis, we propose a binary-mask based noise reduction for the binaural hearing aids. It can preserve the spatial cues which are important for binaural hearing of the human. Besides, it utilizes the concept of the ideal binary mask (IdBM) to improve the speech intelligibility. The proposed binaural noise reduction algorithm not only improves the speech quality as the previous noise reduction but also enhances the speech intelligibility. For the sound localization, it keeps the binaural cues of the input and the output signal the same. From the simulation result, the quality of the speech (SNR improvement) is not worse than other noise reduction algorithms. Moreover, the speech intelligibility (SII) is also enhanced.