Page  287 ï~~Real-time Granular Synthesis on a Distributed Multi-processor Platform Takebumi ITAGAKI #+ Takebumi.itagaki Peter D. MANNING # P.D.Manning Alan PURVIS + Alan.Purvis @ Durham Music Technology* www home page URL=http: // # Department of Music, University of Durham Palace Green, DURHAM DH1 3RL, UK + School of Engineering, University of Durham South Road, DURHAM DH1 3LE, UK Abstract This project is a part of our on-going research into real-time audio synthesis using a distributed parallel processing plat form. In this paper we present our successful implementation of a 9-voice real-time granular synthesis on a part of the 160 Transputer network. Granulation and time-stretch of recorded natural sound are described. 1. Introduction Granular synthesis or sound granulation is a construction of complex sound from grains that were originally proposed as a representation of "acoustic quanta" by a British scientist Gabor in 1947. Granular synthesis is attractive to composers because of its potential for elaborate timbral transformations of sampled sounds and its conceptual simplicity: small fragments of sounds are superimposed to construct more complex sound material. Various composers have explored this technique since the early 1970's. Several software programs are now available for non-real-time granular synthesis, notably CSOUND [Lee 1995]. Due to the nature of the synthesis method, composers generally prefer to interact with the control parameters, thus requiring a real-time implementation. Despite the conceptual simplicity, because of tough requirements for implementing it in real-time most of the working systems have been designed for custom-designed hardware; such as [Bartoo et al. 1994], or a purpose-built "music" workstation; such as NeXT + ISPW and MARS [de Tintis 1995]. * Durham Music Technology Group is a collaboration between the School of Engineering and the Department of Music at the University of Durham. Although high-speed PCs of recent manufacture have some capability for real-time granular synthesis, they still do not have adequate computational capability to meet all demands of a truly interactive composing environment. In this paper we describe an application of realtime granular synthesis for a multi-processor network which achieves the required throughput by means of a specially configured parallel and distributed architecture. 2. The 160 Transputer Network Since 1988 this research group has presented a series of paper concerning multi-processor based audio processing. A prototype module for a transputer network, using 16 T800 transputers, was presented and demonstrated at 1990 ICMC in Glasgow [Bailey et al. 1990]. Subsequently, 10 of these modules were linked together to create a network that is capable of 1400 MIPS. We have reported an application of real-time additive synthesis for the network [Itagaki et al. 1994] and its optimised variations using a dynamic scheduling and a multi-sampling-rate approach [Itagaki et al. 1995a]. As a part of the on-going project on the distributed parallel processing platform, we have implemented a 9 -voice real-time granular synthesis of natural sound. ICMC Proceedings 1996 287 Itagaki et al.

Page  288 ï~~3. Implementation We have configured and programmed a 9-voice real-time granular synthesis system on a part of the 160 transputer network, using 53 networked T800s (17.5 MHz clocked) with 4k-byte on-chip memory, together with 12 external T800s (20 MHz clocked) with 256k-byte memory. The former is mainly used for routing of the control information and the sound output, 25 processors being reserved for sound granulation and address generation. The latter are used for sample storage, capable of storing 128k samples each in 16-bit integer format, a further two processors being assigned for a sound buffer and another for the DAC driver. The granulation parameters (grain speed, grain size, randomise range and time-stretch ratio) are controlled from a host PC via its standard keyboard. Some of the parameters, such as amplitude, can alternatively be controlled using a MIDI keyboard and controller, see [Itagaki et al. 1995b]. The current programme is based on a 20 msec (or 640 samples at a 32 kHz sampling rate) grain which consists of a 160 sample-long rising ramp, a 320 sample-long grain body and a 160 sample-long decaying ramp. The lengths of both ramps are fixed, but, their content could be replaced with either a half cosine curve or a Gaussian characteristic. The speed of grain is variable from 0.1 to 50 grain-per-second (gps) in each voice. The maximum speed is defined as no interval between the grains. The length of the grain body is variable from 0 to 3200 sample-long: this creates a variety of body-ramp ratios. Using a pseudo-random number generator, the interval of the grain and the length of the grain body are randomised within a range defined as a control parameter. A change of parameters can only occur at the end of each grain, therefore, when the grain speed is low, such as 5 gps, the control signal will be updated 5 times a second. As we have experienced in the additive synthesis programmes [Itagaki et al. 1994], the network is capable of handling a sound sample stream at 32 kHz sampling rate with a contra flow of control signal about 500 bytes a second without failure. In case of the granular system, the maximum traffic of the control information will be about 50 gps per voice, which represents significantly lower that possible with additive synthesis programmes. 4. Summary A 9-voice real-time asynchronous granular synthesis programme has been implemented onto a part of the 160 transputer network. The programme is capable of both time-stretching and time-compression of natural sound and has been tested for its reliability. 5. Acknowledgement The authors acknowledge the generous donation of processors from SGS Thomson / INMOS Ltd., UK and financial support from the University of Durham. References [Bailey et al. 1990] Bailey, N.J., Bowler, I., Purvis, A. and Manning, P.D. "An Highly Parallel Architecture for Real-time Music Synthesis and Digital Signal Processing." In Proceedings of ICMC 1990, Glasgow, UK, pp. 169-171 [Bartoo et al. 1994] Bartoo, T., Murphy, D., Ovans, R. and Truax, B. "Granulation and Time-Shifting of Sampled Sound in Realtime with a Quad DSP Audio Computer System." In Proceedings of ICMC 1994, Arhus, DENMARK, pp. 335-337 [Itagaki et al. 1995a] Itagaki, T., Manning, P.D. and Purvis, A. "An Implementation of Optimised Methods for Real-time Sound Synthesis on a Multi-processor Network." In Books of Abstracts Parallel Computing 1995, Gent, BELGIUM, p. 100. [Itagaki et al. 1995b] Itagaki, T., Manning, P.D. and Purvis, A. "An Implementation of Realtime Granular Synthesis on a Multiprocessor Network." In Proceedings of ICMC 1995, Banff, CANADA, pp. 493-494 [Lee 1995] Lee, A.S.C. "CSOUND Granular Synthesis Unit Generator." In Proceedings of ICMC 1995, Banif, CANADA, pp. 230-231 [de Tintis 1995] de Tintis, R. 1995. "GRAINS: a software for real-time granular synthesis and sampling running on the IRIS-MARS workstation." In Proceedings of X1 Colloquic di Informatica Musicale, Bologna, ITALY, pp. 221-223 Itagaki et al. 288 ICMC Proceedings 1996