本研究提出以頻帶分析為基礎的噪音抑制演算法，分別結合離散餘弦轉換(Discrete Cosine Transform)與離散傅立葉轉換(Discrete Fourier Transform)為核心架構實現的近似Class-2、Class-2、以及Class-0 ANSI S1.11 1/3 八度音濾波器組。核心主要之技術為： (1)改善各頻帶間的不匹配現象，亦希望能獲得較低的群延遲與較低的計算複雜度，以及能有效降低聽力圖上的匹配誤差；(2)利用交互相關性為基準的聲源定位演算法，來避免音源能量過強導致判斷上的錯誤；(3)搭配麥克風通道分別計算到的功率頻譜密度分析後進行感知決策；(4)並提出以滑動式/跳點式離散傅立葉轉換即時時頻分析器，做為日後應用於噪音抑制演算法之基礎。相關之技術研發有助於高規格助聽輔具的設計與發展。 This research mainly focuses on developing frequency-subband based noise reduction algorithm, and it integrates with quasi-class-2, class-2, and class-0 ANSI S1.11 1/3-octave filterbank design based on the DCT and DFT modulations, respectively. The kernel techniques involves: (1) Improving the delay mismatch of each subband, and achieving shorter group delay, lower computational complexity, and lower matching error of the audiogram; (2) Using Cross-correlation to realize sound-source localization and avoiding the problem of more sound-source energy more decision error; (3) Calculating and analyzing the power spectrum density of every microphone, and then accomplishing awareness-computation strategy; (4) Propose sliding DFT and hopping DFT computation to obtain the time-frequency information in real time. This research would bring more design concepts and application on the topic of hearing aids, and the proposed method would be more suitable for high-level applications of hearing aids in the future.
