空間響應模擬器的設計與實現及Up/Downmixer

Abstract

本論文的研究著重於如何設計與實現不同空間的空間響應。空間響應模擬是3D音效中相當重要的一環，實現方式可大致分為兩類，一種是將實際量測的房間響應以有限脈衝響應(FIR)的形式對輸入的音訊做摺積(Convolution)，不但耗計算量且成本高。另外一種方式是使用一些無限脈衝響應(IIR)濾波器，經由串聯或並聯等的結構，來近似所要模擬房間的主要特性，可使得運算量大幅下降，然而不適當的係數會造成金屬聲以及一些不自然的聲音。為了兼顧音質以及運算量的考量，本論文提供三種方法來模擬空間的初期反射(early reflections)，利用一系列的IIR濾波器來產生晚期混響(late reverberation)。我們建立了一套以基因演算法(GA)為基礎的最佳化架構來最佳化空間響應模擬器的參數。進行了主觀以及客觀的實驗評估，最後經由實驗結果可提供兼顧了音質與運算量的空間響應模擬器設計流程，並且將空間響應模擬器應用於Up/downmix，以提供最佳的三度空間立體音效。

This thesis focuses on how to design and implement the room response simulators of different rooms. Room response simulator is an important element in three-dimensional audio reproduction. It can make music sounded more spaciously. The ways to implement can be generally separated into two kinds of approaches. One approach is using the measured real room impulse response as a FIR filter. And then convolutes with input audio signals. High complexity of computation and large memory are needed. Another approach is using the series connection or parallel connection of some IIR filters to approximate the principal characters of desired room response. The complexity of computation will substantially cut down. However, it’s easy to produce metallic and other artifacts with improperly designed room response simulators. In view of the trade-off of complexity and performance, three methods are proposed to simulate early reflections and the late reverberation is simulated by using the IIR filter network structure. Series of steps are established to optimize the parameters of room response simulators based on genetic algorithm (GA). Subjective and objective experimental evaluations are conducted to find out the design guideline of room response simulators. And this technique is as an application of up/downmix processing to provide excellent performance of three-dimensional audio effect.