頭部轉換函數在共極/零點模式下之研究

Abstract

HRTF資料集合為一組由聲源至雙耳的脈衝響應，它廣泛地被使用在3 - D音訊系統，然而，大量的HRTF資料量造成即時處理上的困難，因此，縮減龐大的HRTF資料量一直是3 - D音訊處理的中心課題。 由於耳翼具有共振腔的特性，一般認為，不同聲源的HRTF可能具有共同的共振頻率，因此，在本論文中，吾人採用共極╱零點模式於HRTF資料群來取代原本每一HRTF即採用一極╱零點模式的作法。共極點及零點的決定將基於Prony, Shanks, 及iterative prefiltering三種方法，此外，HRTF的分群也將在套用共極╱零點模式前進行，以便能降低套用模式造成的誤差。 最後，藉由HRTF的合成將驗證以上的方法，同時，聆聽測試阤將被進行來證實其在人類聽覺上的可行性。

HRTFs are impulse responses from sound source to both ears. They are widely used in 3 - D sound applications. However, the huge amount of HRTFs data set makes real - time processing diffcult. Therefore, reducing the huge HRTFs data has been an central issue on 3 - D sound processing. Due to the resonant characteristic of pinna structure, it is believed that the HRTFs at different eosiions may share common resonant frequencies. In this thesis, thus, we adopt the common - pole / zero model individually. The determination of fitting every HRTF by a pole / zero model to fit the HRTFs data set instead of fitting every HRTF by a pole / zero model individually. The determination of common - poles and zeros will be based on three methods, Prony, Shanks, and iterative prefiltering. Besides, clustering of HRTFs is made before applying common - pole / zero model to lower the model error. The simulations of HRTFs synthesis will be demonstrated to verify these approaches. Listening tests are also made to justify the validity in human hearing.