Page  477 ï~~Musical Instrument Transient Analysis and Re-Synthesis Anna I. Katrami, Lecturer in Electronic Music EIMC Unit, University of Bradford, Bradford, BD7-1DP, U.K. Tel. +44-1274-385467, Fax: +44-1274-383727 e-mail: http://www. Dr. Peter J. Comerford, MMR Unit Computing Department University of Bradford Bradford, BD7-1DP, U.K. Tel. +44-1274-383936 e-mail: Abstract The aim of this paper is to describe a reduction process for transient analysis data of pipe organ sounds, in order to minimise the information required for resynthesising them without significant perceived change on the sound. 1. Introduction For some considerable time musicians and researchers have been analysing acoustic instrument sounds in order to find the individual characteristics that distinguish one from the other. One of the main components which determines the way we categorise different sounds is the attack transient, in other words the way the amplitude and the frequency of each harmonic component changes according to time. 2. Attack Transients a Figure 1 Example of attack transients for different partials An attack transient is a period of extreme spectral disorder, since partials change in frequency, attack at different rates, and overshoot their sustain-state amplitudes in different ways (see Figure 1). Prominent unpitched wind noise can also be present. This disorder and unpredictability of the attack transients is very significant in establishing note identity [1] and has a profound effect upon the way that sounds are perceived since the envelopes and duration's differ both between different instruments and between different notes on the same instrument [2]. in order to simulate any instrument sound on a computer system it is necessary to program all the individual characteristics required to generate the sound with as little perceived difference as possible from the original. At same time it is important to save computer memory and running time. This has motivated us to create an environment which will combine both aspects and will focus mainly on the organ sounds. 3. Phychoacoustic Tests Our work also includes psychoacoustic tests carried out to measure perceived differences between original recordings and resynthesised tones. ICMC Proceedings 1996 477 Katrami & Comerford

Page  478 ï~~The tests used are based on 'discrimination'. In this way listeners are presented with a pair of sounds every time that are identical or different in some way. The first sound is the original and the second is the re-synthesised tone with the appropriate modifications i order to reduce the amount of data necessary for simulating the attack transient [3]. The degree of difference between the two sounds presented varies in order to understand the listener's sensitivity to the amount of change along a given stimulus dimension. Data reduction in the synthesised waveforms is obtained by using: " common amplitude attack envelopes for different partials, " common frequency attack envelopes for the different partials, " common initialisation times for the partials, " simplified envelope shapes, " additional noise. 4. Tools The Analysis I Synthesis technique used to create the sound examples is based on the Phase Vocoder [4] which: " produces a time-varying spectra, using a series of FFTs, and " can reliably re-synthesise modified tones without any loss of information. The graphics representation of the way the amplitude and frequency components are changing according to time is produced using Matlab. 5. Summary As we described above, attack transients are very important in order to categorise different instrument sounds. By changing the frequency envelope of the partials, the attack rates, the overshoot position, and the additional noise (which is very important for pipe organ tones), the perception of the sound changes completely. The psychoacoustic tests we carried out were based on the principal of defining what are the most important characteristics of the attack transients when we try to simulate a pipe organ sound'. References [11 L.K. Comerford, P.J. Conerford & A.I Katrami, 'Synthesis of audio waveforms for the measurement of perceived organ ensemble in varied acoustics', Proceedings of Institute of Acoustics 11th Annual Conference, Windermere, UK, November 16-19, 1995, Institute of Acoustics Journal, Vol. 17, Part 7 (1995), pp. 147-158. [21 Stephen Handel, 'Listening: An Introduction to the perception of Auditory Events', MIT Press, 1989. [31 John AM. Grey & James A. Moorer 'Perceptual evaluations of synthesised musical instrument tones', Journal of Acoustical Society of America, Vol. 62, No. 2, August 1977. [41 John Stralwn, 'Digital Audio Signal Processing: An Anthology', William Kaufinann, Inc., Los Altos, California, 1985. ' Examples of the tests carried out will be presented in the conference. Katrami & Comerford 478 ICMC Proceedings 1996