1 Speaker Auralization – Subjective Evaluation of Nonlinear Distortion Wolfgang Klippel Klippel GmbH Germany www.klippel.de ABSTRACT A new auralization technique is presented for the objective and subjective assessment of drivers in the large signal domain. Using the results of the large signal parameter identification a digital model of the particular driver is realized in a digital signal processor (DSP) to simulate the sound pressure output for any given input signal (test signal, music ). This technique combines objective analysis and subjective listening test to assess the linear and distortion components in real time. This valuable tool shows the impact of each distortion component on sound quality and allows driver optimization with respect to performance, size, weight and cost. Introduction Loudspeakers which have similar small signal characteristics may sound quite different at higher amplitudes. Thermal and nonlinear mechanisms determine the maximal output of the driver and cause signal distortion. Most common measurement techniques valuate the performance at minor amplitudes. Assessing the driver in the large signal domain says more about the performance of the speaker under normal working conditions. Loudspeaker are expected to be able to reproduce the sound as loudly as possible with low distortion. Cost, sensitivity, size and weight are other constraints which are directly related to the large signal performance. There are different objective approaches to measure the performance objectively. Harmonic and intermodulation distortion measurement show the effect of the nonlinearities for a special excitation signal. Nonlinear and thermal parameters show the physical causes of the distortion and are crucial for the driver design. However, the results of both measurements do not reveal the impact of the distortion on subjectively perceived sound quality. There are many other questions in the gap between subjective and objective assessment: • How sounds the distortion components produced by the separated nonlinearities? • How can we mask distortion to make them less audible? • What is the most critical program material in listening tests? • Are there any desired effects of nonlinear distortion? • How can the performance/cost ratio be increased? • What are the benefits of some additional efforts? • How does the driver sound after optimization? In order to give answers to this question we will present a new auralization technique which combines objective distortion measurements on audio signals in real time with a tool for performing systematic listening tests in the large signal domain. After giving a summary on large signal modeling we will apply this technique to artificial and natural sounds and will discuss the relationship between objective parameters and subjective sensations. Loudspeaker Modeling M ms C ms (x) R ms Bl(x) L e (x) R e (T V ) v F m (x,i) i Bl(x)v Bl(x)i u Fig. 1. Large signal model of the driver At low frequencies the driver may be modeled by a lumped parameter model comprising electrical and mechanical elements and state variables: u voltage at terminal i electrical input current x voice coil displacement v voice coil velocity (dx/dt) Fm(x,i) reluctance force TV voice coil temperature Re(TV) electric DC resistance depending on voice coil temperature L e (x) voice coil inductance Bl(x) electrodynamic force factor (Bl-product) Kms(x) mechanical stiffness of driver suspension Mms moving mass including air load R ms resistance representing mechanical and acoustical losses To model the dominant nonlinearities of the driver at high amplitudes this model considers the variation of the force factor Bl(x), the stiffness K ms (x), and the inductance L e (x). Additional parameters L 2 (x) and R2(x) used for modeling para-inductance at higher frequencies are omitted here in this paper to keep the following equation as simple as possible. The electromechanical equivalent circuit corresponds with the following set of nonlinear differential equations: dt i x L d dt dx x Bl i R u e e ) ) ( ( ) ( + + = x x K dt dx R dt x d M i i dx x dL x Bl ms ms ms e ) ( ) ( ) ( 2 2 + + =       +