Prismatic Synthesis

I've always been fascinated by various forms of sound synthesis. Additive techniques are precise but laborious, while wavetable-based subtractive synthesis

I've always been fascinated by various forms of sound synthesis. Additive techniques are precise but laborious, while wavetable-based subtractive synthesis is much easier to use if you want to paint with a broad sonic brush. Physical modeling provides unprecedented expressive potential, but it's a real bear to program anything useful with that technology.

After what seems like a long dry spell in the development of new synthesis techniques, something interesting is about to emerge from a company called EuPhonics, which was recently acquired by 3Com ( Called Parametric Resynthesis by Interpolated Signal Models (PRISM), this technology vaguely resembles certain aspects of wavetable and physical-modeling synthesis, but it's really something new.

As with sampling, a PRISM sound starts with a recording of one or more notes. These recordings are then separated into pitched and unpitched (noise) components, each of which is analyzed to determine its power envelope and spectral characteristics. A pitch envelope is also derived from the pitched component. The spectral characteristics are represented by a series of filter coefficients, which can vary in time along with the pitch and power envelopes (see Fig. 1).

Once the analysis is complete, the derived parameters are used as a model to reconstruct the original sound and manipulate it in any number of ways. Unlike physical modeling, which is based on complex equations that describe the general acoustic behavior of certain types of instruments, PRISM is based on specific recordings. As a result, this technique is called signal modeling, and the timbre of the synthesized sound depends entirely on the quality of the source recording.

Among the many differences between PRISM and wavetable synthesis is the fact that PRISM sounds are not looped. Instead, the derived parameters allow a note to be held for any duration, regardless of the original recording's length. In addition, sustained notes are not static; all the irregularities of the original human player can be modeled and applied to the synthesized sound. PRISM instruments can also include the natural vibrato, tremolo, and release of the original sound, and the formants do not shift along with the pitch of the notes. All of this contributes to the natural, organic quality of the final sounds.

Even better, the system smoothly interpolates between different sets of parameters, which allows seamless morphing between the timbral characteristics of notes played at soft and loud volumes, at low and high pitches, and even by completely different instruments. This goes way beyond crossfading and provides exceptional expressive potential, especially when you consider that all of these activities (and many more) can be placed under independent MIDI control. In addition, you need to record only one soft note and one loud note per octave to recreate a realistic simulacrum of the original instrument. Try that with a sampler!

Best of all, PRISM sounds will be easy for end users to create. 3Com plans to develop software that automates much of the analysis process and allows musicians to manipulate the results in musically meaningful and intuitive ways. Using such software, any sort of real or imagined acoustic or electronic sound can be easily realized. The first commercial PRISM product is likely to be a PC sound card for the gaming and desktop-audio markets; it should be available by mid-2000.

I've heard some early prototype examples of PRISM sounds, and I was impressed. In particular, the examples I heard included a vocal sound that knocked my socks off. The flute was excellent as well, but the saxes still needed some work. I'm very excited about the potential this technology offers to electronic musicians, and I can't wait to plug my MIDI wind controller into a PRISM synth. Be still my beating heart!