Modern software-based musical instruments tend to be centered around the keyboard paradigm in which sound synthesis is controlled merely by discrete pitch and amplitude values. This paucity of information, however, places severe limits on the kinds of music that can be made with such instruments, and for this reason the term “electronic music” has become synonymous with different styles of dance music. I propose to create a real-time synthesizer that can be controlled by continuous values for pitch, amplitude and brightness. This project would be realized using the CLAM library for audio analysis and synthesis and would further the goals of this project by providing an example application illustrating the use of spectral modeling synthesis. Spectral modeling synthesis is a variety of sound synthesis that is particularly well suited to the stated task, because it creates models of recorded sounds which can be manipulated in interesting ways without noticeable loss of sound quality.

In the first stage of the project, I will create a database of samples to be used by the synthesizer. I will make recordings of an ebow playing a guitar string with pitches at different levels of amplitude and brightness. (An ebow is a device which creates a magnetic field which can be used to make steel strings vibrate.) I will derive spectral models of these recordings using CLAM’s SMSTool, and I will save segments of these recordings with uniform parameter values in SDIF format.

In the second stage I will design the database for managing these SDIF files. It will allow clients to query the database using values for pitch, loudness and brightness and retrieve in real-time the best fitting segment for a given set of query values. Due to the requirement of real-time processing, the database will likely have to make use of either caching or sample pre-loading.

In the third stage I will design the synthesis engine that will be responsible for interpolating between segments and resynthesizing the audio from its spectral representation. Given the nature of the sound source, it is possible that a sinusoidal model will give an adequate representation of the sonic material, and therefore, the complexities of resynthesizing the residual portion of the signal can be avoided. I assume that this would be preferable as one of the goals of this project is to create a simpler, more understandable version of a previous CLAM project named SALTO which performed a full-blown sinusoid plus residual resynthesis of saxophone recordings.