HRTFs的共極/零點模型與分群研究

Abstract

在三度音訊處理中所使用的頭部相關轉換函數(HRTFs)會隨著位置而改變。為了完整描述頭部週遭音訊定位的訊息，大量的資料會增加實做的困難。低階的IIR濾波器是有效建立HRTF模型的方法。為了進一步減少儲存模型參數所需要的記憶體跟節省計算量，我們提出共同平衡模式截斷演算法，來設計具有共同極點的IIR濾波器。在共極/零點模型中，極點並不會隨著音訊位置不同而改變，零點是用來表示出不同HRTF間的差異性。並非所有HRTF擁有共同極點的特性，要強制所有的HRTF共用極點有些牽強 。因此將HRTF分群是非常必要的。將HRTF分成較小的群組，而不同的群組可以擁有各自不同的極點是比較實際的方法。我們提出共同極點 LBG演算法。此演算法利用遞迴式更新組合的方式來找到最佳的分群結果。最後電腦模擬會驗證我們所提出來的方法。我們也會證實我們所提 出的方法比過去的方法具有優勢。

Head-related transfer functions (HRTFs) used in 3-D sound processing are locationally dependent. Fully modeling sound localization cues around the head burdens implementation with a huge data set. Low-order IIR filters are an efficient way to model HRTFs. To further reduce the required memory for storage of model parameters and save computational cost, we propose jointly balanced model truncation for design of common-pole IIR filters. In common-acoustic-poles/zeros (CAPZ) modeling, poles are independent of sound directions and zeros are used to show differences between HRTFs. Not all measured HRTFs have common-pole characteristics, so it is impractical to force all HRTFs share common poles. It is more practical to divide HRTFs into groups and each group has its own common poles. Therefore, clustering of HRTFs becomes essential. We propose an algorithm, common-pole LBG, which iteratively updates the grouping to reach the optimal clustering. Finally, computer simulations are used to verify our proposed methods. We will demonstrate our proposed methods have advantage over previous works.