模糊集類神經音像定位控制及聆聽室聲學研究

Abstract

本論文可分為兩部份：第一部份為聆聽室聲學音場恆定之研究。傳統音響 再生過程中，受到聆聽室的殘響及再生音源等非線性失真等影響，使得聆 聽室中無法精確地重建原錄音環境之音場。為了消除上述之誤差源，我們 提出一經由逆遞迴式類神經網路控制器及正向模型所組構成的適應性控制 機構以便克服高傳真音響系統之非線性失真及模型辨認的困難。第二部份 是研究不同音場對人類聽覺認知系統的影響。進而提出模糊邏輯音像定位 控制系統架構。傳統立體聲音像控制系統中，可藉由控制各聲道的再生強 度及相位差，以達到音像定位的目的。但受到人類聽覺系統認知上的差異 及聆聽環境變動等因素的影響，控制參數和人類音像認知的相關特性無法 精確地描述出來。因而我們提出的適應性模糊控制系統藉由語意變數的描 術及模糊推論法則，可將人類聽覺特性加以估測以達到音像定位之目的。 This thesis includes two parts. The first part describles the design of a novel-based acoustic control used for the equalization of the response of a sound reproduction system. Since, the traditional adaptive equalizer is capable of dealing with linear systems or specific nonlinear systems, the time- delay feedforward neural network (TDNN) which bave the capability to learn arbitrary nonlinearity and process the temporal audio patterns are particularly recognized as the best nonlinear equalization of the sound reproduction. The performance of TDNN-based acoustic controller is verified by some simulation results. The second part presents a new fuzzy logic control (FLC) approach which leads to a sterephononic reproduction controller for localizing an auditory image in the desired direction and distance. The localization blur of auditory is usually less precisely resolved than physical sound space. In other words, it turns out that controlling the auditory image is more difficult than the sound controll. Different from the conventional sound image localization approach, the fuzzy logic controller can take into account human auditory perception knowledge. The ambiguous human auditory perception can be represented by a number of fuzzy-set values. From these fuzzy representations the auditory image localization controller characterizes the function of how control outputs depend on control inputs as fuzzy implications or associations. Furthermore, the overall sterephonic reproduction controller can be realized be a 45-rule fuzzy associative memory FAM system. The performance of FLC-based auditory image localization is verified by a number of experiments.