Вы находитесь на странице: 1из 6

Historic Development of Reverberation Algorithms

Brian Tuohy 40068747

MUS7080 Signals, Sounds & Senses 29th November 2011

Artificial Reverb allows us to apply different spatial characteristics to a given audio signal (Moorer 1979). Computers can digitally add reverb to dry inputs, allowing for control of delay time, spectral content and decay characteristics. The computer outputs several signals, which attempt to recreate the reverberant characteristics of real environments. Manfred Schroeder was among the first to publish computer simulations of room reverberation. As opposed to previous artificial reverberation methods, Schroeder's paper (1961) proposes a filtering method that does not affect the amplitude spectrum. This was achieved through the use of unit reverberators recursive comb filters and all-pass filters. Schroeder suggested a reverb model that used five all-pass filters in series added to a scaled portion of the original signal. Another model consisted of four comb filters in parallel, fed into two all-pass filters in series, and then added to a certain portion of original input. These filters essentially simulate the effect of wall reflections and the decay of sound over time (Moorer 1979). The mix between the direct sound and reverberant sound simulates the distance between the listener and the original sound source. One of the significant improvements afforded by Schroeder's reverberators was the avoidance of the hollow sound associated with earlier filter designs due to the flat frequency response exhibited by the all-pass filters. Some early digital reverberators had a very low echo density, producing as few as 30 echos per second, when real concert halls would produce more than 1000 per second (Roads 1996). Early reverberation algorithms also created equally spaced impulses, whereas real rooms have a more diffuse sound energy between early reflections (Toma, Topa & Szopos 2005). These problems were overcome by using all-pass filters. Connecting all-pass filters in series essentially multiplies the number of impulses in any given time (Griesinger 1989). Hence, by connecting all-pass filters in series, Schroeders design created high-echo density and resulted in smooth response. These filters provide exponential decay with time and flat frequency response, which avoids the undesirable metallic quality exhibited by other artificial reverberators that use delay and feedback. Phase response of all-pass filters, however, can be very complex they are not entirely colorless, as the filter can ring with a frequency equal to its delay time if a short transient sound is applied (Roads 1996). Schroeders reverb models are useful for global reverb, but not for simulating specific acoustical responses of actual venues (Roads 1996). Extensions to the Schroeder model include replacing the all-pass filter with an oscillatory all-pass filter. This helps to create a more realistic room sound with a reverberation that fluctuates slightly. We now know that the decay rate of reverberation is dependent on the shape of 1

the space and the use of sound absorbing materials (Schroeder 1970). Considering these factors, the behavior of sound waves and reverberation within an enclosure can be estimated by geometrical means. FR Moore presented geometric room modelling methods which attempted to estimate the path of sound reflections from sources in a virtual room (Roads 1996). This kind of computer ray tracing was found to be much more accurate than traditional formulae for decay time calculation such as those proposed by Eyring, Sabine and Millington (Schroeder 1970). However, much of this type of modelling fails to account for diffusion, whereby the sound rays are scattered and partially absorbed due to the presence of surfaces which are not entirely flat or reflective (Roads 1996). Each time a ray hits an absorbent material, its energy is reduced by (1-a) where a = the absorption coefficient. The absorption is also dependent on angle of incidence (Schroeder 1970) Moorer designed modifications of Schroeder's reverberators to account for similar discrepancies (Moorer 1979). It was suggested that placing a low-pass filter in the loop could simulate the attenuation of higher frequency sounds in air. Such attenuation depends on temperature, humidity, pressure and frequency. These attenuation values, although generally quite low, can have a very significant effect on the spectral content of reverberated sound in large venues, where there can be a significant difference between the amplitudes of high and low frequencies in terms of reverberation and attenuation. With Moorers design, a simple one-pole filter is added to act as a low-pass filter, which simulated the absorption of high frequencies in air. Moorers modification of Schroeders design suggested six comb filters, each with a low-pass filter in the loop, and a single all-pass filter. This is only a vague approximation of absorption in air. Schroeder also suggested estimating the reverberant qualities of a room by measuring the response of a single pulse (impulse response) (Schroeder 1970). This response could then be simulated and convolved with another sound, essentially representing that sound in the modeled room. Direct convolution, however, is not practical for reverberation due to its huge computational demands per second of audio (138:1 for a stereo sample) (Roads 1996). This method of generating reverb also lacks the parametric control that may be desirable for an interactive system (Gardner 2002). These problems presented the need for an adapted method of convolution. Early reflections account for the first 40-80 milliseconds of decay (Moorer 1979). These early reflections can be passed through the impulse response image and the later echoes can be modeled with a modified Schroeder reverberator. The convolved early response feeds into the Schroeder reverberator and the echoes coincide, creating the source for new reverberation and reducing the 2

computational load. This way, we can accurately represent the early reflections of a real venue with a multi-tap delay line (tapped recirculating delay - TRD), and use this to feed a relatively computationally inexpensive reverberator that produces the fused reverberant sound (Roads 1996). Moorer (1970) proposed such a design with a 19 tap finite impulse response model feeding into a circulating reverb model with 6 comb filters in parallel, each with an individual low-pass filter, all summed and fed into a single all-pass filter network with gain of about 0.7 and delay of about 6 ms. The model proposed here overcomes fluttery response to transients and metallic decays experienced with some previous reverb systems These simulation methods can be used to propose new architectural designs for theaters and concert halls or to predict the results of changes to existing structures (Schroeder 1970). This also allows us to evaluate the perceived effect of different environments on a musical piece; how it might sound if performed in a particular location, without having to go to the location and attempt to determine what factors may be contributing to a desirable/undesirable sound. Also important in the simulation of accurate reverb is the decorrelation of the reverb signals sent to each ear in order to create a spatialized effect (Roads 1996). Griesinger addressed this issue and took a similar approach to Moorer with his Lexicon delay algorithm (Toma, Topa & Szopos 2005). This algorithm presented the direct sound and 3 lateral reflections on each side of a stereo output, with delay times between 10 and 80 ms. Each set passes through an all-pass filter and a low-pass filter at 2 KHz. Griesingers most successful system for reverberation contains an early binaural reverberator, all-pass filters, and absorbent an all-pass late reverberator. This produces results much closer to those of a real room. This kind of reverberator offers advantages such as accurate modeling of early reverb, increased impulse density with time for late reverberation, uncorrelated output, a relationship between reverb time and frequency, and elimination of metallic sounds. Reverb is also used for more creative means, rather than simply modelling spaces. With digital reverberation, it is possible to create fictional spaces that could not be reproduced in the real world (Roads 1996). More recent algorithms have largely been based upon the designs of Schroeder but have become mathematically much more complex and permit techniques that are not always intent on realistic simulation, but rather an artistic application of complex spatial transform methods.

Granular reverberation can use a cloud of sound grains to convolve around the original sound, smearing its signal and creating a reverb effect (Roads 1996). Multiple stream reverberation splits the reverberated signal into multiple streams which each model the reverb from a small space in a virtual room. A TRD network is adapted to suit that location in the room, and all of the streams are then summed together to represent the overall reverberation of the space. Digital simulation of reverb is important for answering theoretical questions relating to sound transmission and reverberation in the real world. Such an understanding is important for the study of perception of physical sound and reverberation and how it is affected by physical factors in the environment (Schroeder 1970). Although digital reverberation algorithms are very advanced, many still hold the belief that the best acoustic environments are still superior to any artificial representations. What artificial reverb does provide, though, is flexibility (Roads 1996). By using relatively computationally inexpensive methods, we can create fictional environments that could never be actualized, or switch between entirely different reverberant responses within a single section of music.

Bibliography Gardner, W. 2002, Reverberation Algorithms in Applications of Digital Signal Processing to Audio and Acoustics, eds. M. Kahrs & K. Brandenburg, Springer US, pp. 85-131. Griesinger, D. 1989, Practical Processors and Programs for Digital Reverberation, Audio Engineering Society Conference: 7th International Conference: Audio in Digital Times, 5. Jot, J. 1997, Efficient models for reverberation and distance rendering in computer music and virtual audio reality, Physical Modeling II, The International Computer Music Association (ICMA), Aristotle University of Thessaloniki, Hellas, 25-30 September 1997, pp. 236. Moorer, J.A. 1979, About this reverberation business, Computer Music Tutorial, vol. 23, no. 2, pp. 12 - 28. Roads, C. 1996, The computer music tutorial, MIT Press. Schroeder, M.R. 1961, Improved Quasi-Stereophony and Colorless Artificial Reverberation, Journal of the Acoustical Society of America, vol. 33, no. 8, pp. 1061 - 1064. Schroeder, M.R. 1970, Digital simulation of sound transmission in reverberant spaces, The Journal of the Acoustical Society of America, vol. 47, pp. 424. Toma, N., Topa, M. & Szopos, E. 2005, Aspects of Reverberation Algorithms, Proceedings of the International Symposium on Signals, Circuits and Systems, pp. 577-580.