Subject: Re: Reverberation time From: Christian Borss <christian.borss@xxxxxxxx> Date: Mon, 20 Jul 2009 10:23:50 +0200 List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>Am 19. Jul. 2009 um 17:46 Uhr schrieb Zhao LiHeng: > Recently, I was reading papers about speech separation in the > reverberant environment. Generally speaking, most of the previous > works are carried on simulated data, in which the reverberant speech > is the convolution of a clean speech and the simulated room impulse > response provided by some toolkits, such as Roomsim (given by D. R. > Campbell). However, it is surprising to me that the reverberation time > can be set to a certain value (e.g. RT60 = 300ms) in some papers. As I > know, reverberation time is influenced by many factors, such as room > volume, absorption coefficient of the surfaces et al. Therefore, I > have the following questions: > 1. How to set the reverberation time to be a certain value in a > simulated room? Is there any paper or toolkit for performing it? As Densil Cabrera already stated, you can start with Sabine's formula to obtain reflection coefficients which result in a desired reverberation time, H(f) = sqrt( 1 - 0.163 * V / (T_60(f) * S) ) , where V is the volume and S is the surface of the room. So what you obviously need besides the frequency-dependent reverberation time, is the room geometry. That's simply what I'm using in my auralization software (tinyAVE) for rendering early reflections: http://www.ruhr-uni-bochum.de/ika/ika/forschung/gruppe_martin/hum_mach_interf/aud_virtual_environment/tinyAVE_engl.htm For rendering the late reverberation tail, a statistical model is more appropriate than using an image source model with an insane high reflection order, because in real rooms you don't have this regular specular reflection pattern. Especially, if you want to use the synthesized room impulse responses for some kind of statistical signal processing approaches, it is questionable, if these deterministic signals are appropriate. Whenever my colleagues or our students want to use a simple shoebox image source model for signal enhancement, I advice them to use real measurements for this reason. > 2. If the other factors are invariable, how much will the > reverberation time change according with the location variances > of source or microphone? Is the reverberation time mainly > determined by room configuration? Depends on the room. There are good-natured rooms where the invariance assumption holds and there are ill-natured. :-/ Regards, Christian -- Christian Borß, Dipl.-Ing. || Institut für Kommunikationsakustik http://www.ika.ruhr-uni-bochum.de || Ruhr-Universität Bochum Tel.: +49-(0)234-32-22470 || Universitätsstr. 150, IC1/33 Fax.: +49-(0)234-32-14165 || D-44780 Bochum (Germany)